Rene Haas in Conversation with Top Leaders

Rene Haas [0:00]
Welcome to Tech Unheard, the podcast that takes you behind the scenes of the most exciting developments in technology. I’m Rene Haas, your host and CEO of Arm. Today, I’m joined by Jayshree Ullal, president and CEO of computer networking company Arista Networks. Jayshree spent her early career working in semiconductors at Fairchild Semiconductor and Advanced Micro Devices, then spent a decade and a half at Cisco where she led their data center, switching, and services groups, before going on to lead one of America’s foremost tech companies.

Rene Haas [0:37]
Jayshree. Welcome to Arm. Welcome to Tech Unheard. Thank you for coming all this way.

Jayshree Ullal [0:41]
Well, it wasn't that far. It's a mile away. But thank you for having me.

Rene Haas [0:45]
We didn't know where exactly you had come from, so only a mile away. So very happy to have you.

Jayshree Ullal [0:48]
I could have walked on a good day.

Rene Haas [0:50]
And today is a nice day.

Jayshree Ullal [0:51]
Gorgeous. Yeah.

Rene Haas [0:52]
Thank you so much for joining us. You and I have not known each other super long, but in some ways we have, in terms of when we think about our years in the industry and where our paths cross. And one of the things I love to do to kick these things off is just have the guest tell me a little bit about their upbringing and not only where you grew up, but how you kind of got into the world of tech.

Jayshree Ullal [1:10]
Yeah, no, absolutely. Let me start at the very beginning. I was born in London in Paddington General Hospital, so my early years were very much shaped by British influence. Even today, I like a good cup of tea and cucumber sandwiches and I like to watch cricket when time permits, at least the highlights. But I was raised mostly in New Delhi, India, so a lot of my, you know, formative years was in India where I went to an all-girls Catholic school.

Rene Haas [1:36]
When did you leave the UK? What age?

Jayshree Ullal [1:37]
About four or five years of age.

Rene Haas [1:38]
Do you remember the UK?

Jayshree Ullal [1:39]
I do remember, because my younger sister was born and I actually remember the cruise – my dad chose to go by ship rather than by plane. So it's 21 days. So I remember that well and including going through the Suez Canal, and it was a long journey which ended up in the port of Bombay. And my late aunt was so excited to see me because none of them had seen me since birth. She tried to lift me and she couldn't. So that's a different story. My parents certainly made sure that I was well-fed.

Rene Haas [2:12]
So Delhi. You went to, you said an all-girls school?

Jayshree Ullal [2:16]
Yes. Yes. In those days, and I think it's still true in India, that the best education is the private schools that are typically run by systems defined abroad. So it was run by Catholic Irish nuns. You know, my parents are Hindu, so I had a lot of influence to the Hindu culture. But I also got to experience a lot of the Catholic culture, you know, and Christmas was a great holiday spirit. We learned all the songs and we knew all that good stuff. So it was nice. It was a nice secular upbringing, I would say.

Rene Haas [2:44]
So what started to get your slant towards engineering and tech? When did that happen?

Jayshree Ullal [2:49]
It probably happened much later, you know, in the high school years, if you say, because I was there from kindergarten through graduation. And what happens is, you are asked to kind of fork in the road – are you in humanities or are you in science, is kind of the teachers’ pose to you. And mine was through a process of elimination because I wasn't good at a lot of the humanities subjects, I liked math and physics, although the biology dissections of the frogs left much to be desired. But they make you actually take a board exam where you decide which way you go. And in my case, it wasn't clear cut that I was absolutely good at science or terrible at humanities.

Rene Haas [3:30]
And this is what, like grade nine, ten equivalent?

Jayshree Ullal [3:32]
Yea, grade nine is probably your teenage years. Out of the 200 students, 150 end up in humanities sections B, C, D, and section A is the science section of smart, smart kids. So there's a little danger in that because all of a sudden all the smart kids are in one section and, you know, you thought you were smart, but then so is everyone else. But through a process of elimination, I think I found myself in that field because the other ones, I wasn't as good at – history, geography, needlework, foreign languages. Those were all not my strengths, so I veered to my strengths.

Rene Haas [4:06]
But did you have an interest in science and engineering, electronics?

Jayshree Ullal [4:10]
I did, not engineering and electronics so much. But I had more of an interest in pure science.

Rene Haas [4:14]
Pure sciences, yeah.

Jayshree Ullal [4:15]
Because engineering and electronics were not fields in high school, right? But if you are good at physics and you’re good at math and you're good at chemistry. And in fact, had I stayed in India, I would have probably gone towards a pure science route.

Rene Haas [4:28]
So what – after that period, to your university years, how did you make that choice?

Jayshree Ullal [4:33]
Yeah. So my dad had started the first five IITs in India, which is an incredible, huge deal. So, and he always said, ‘oh, you got to look at engineering’, but as long as dads are involved with it, you don't think about it as much, right, but when I came to this country, it became clear that applying my physics or math, aka engineering, would be a much easier and much more applicable route than a pure science. So I ended up in electrical engineering and actually back in my time, your time too probably, computer science wasn't even offered as a field. I really wanted to do computer science. So I ended up going to the math department to take some computer science classes and enjoying it, and doing well.

Rene Haas [5:14]
Yeah, without belying our ages, you know, at that time, electro engineering was electromagnetic fields and waves, power distribution.

Jayshree Ullal [5:20]
Exactly, thermodynamics. A lot of analog to digital. I still remember this professor, Italian professor, Franco. One day we all failed a test because he had given us these sort of problems that were thinking problems. And then he said, ‘this duster eraser has more brains than you guys do and you're not going to make it in real world’. And it's nice to see Dr. Franco’s proud of me now because he says, ‘you know, out of all of them you had the potential’, so he’s backtracked a little.

Rene Haas [5:47]
For me, our university offered computer engineering as a major, which was this mix between hardware and software–

Jayshree Ullal [5:56]
That’s great.

Rene Haas [5:57]
– which I really got into. And we did programming on ImpSci 80-80s and things of that nature. What were you using back in the day?

Jayshree Ullal [6:04]
Me too, I did some programming on the Z80 and the X86, and I actually got some – I used to come driving to the Silicon Valley to get some samples of chips that had been failed or flawed, because obviously the good ones were sold. That was more of the assembly programming, as you remember. And then we also did some Fortran high-level programming where I had these batch cards. Did you know about batch cards? Yeah. Okay. So those kind of programming was challenging because you'd go there and get all of it done. And then, you know, I was a little careless and ended up dropping the cards and that was like a moment of, ‘oh my God, I got to reprogram the whole thing again’.

Rene Haas [6:42]
Batch cards taught you to be precise and be very, very efficient. So I was always envious of kids like you who got to be in Silicon Valley and to go down the street to Fry’s and actually buy Z80s.

Jayshree Ullal [6:53]
We did, I mean, in fact, right here, one of the companies I worked at, AMD, Fry’s was literally a tourism store, right? Anybody who came would go to Fry’s, right, and it was right in our backyard.

Rene Haas [7:04]
Fries and there was this place called Weird Stuff Electronics in Sunnyvale. I would always be so jealous of you guys. So graduate in Silicon Valley. What was your first job?

Jayshree Ullal [7:11]
Well, first I thought I'd go back with my parents. And this was just, you know, a short stint to get my undergrad done. But life had different plans. And, you know, I interviewed and ended up in a company called Fairchild Semiconductor, who doesn't exist anymore. I think it's part of OnSemi.

Rene Haas [7:26]
Fairchild is the grandfather of all semicos.

Jayshree Ullal [7:29]
That's right. A lot of the people I worked with and I trained with, certainly – it had just gotten bought by Schlumberger. And I was, you know, a product engineer and I had to come into Milpitas to the design factory and do a lot of the memory chip product and design.

Rene Haas [7:43]
They had products?

Jayshree Ullal [7:44]
These are memory chips. These are 4K static RAMs that went into the ECL, not even CMOS. Bipolar. That's right.

Rene Haas [7:52]
ECL. I don't think I knew that. I was a product engineer too.

Jayshree Ullal [7:55]
I love that field.

Rene Haas [7:57]
At TI, on memory. 64K DRAM.

Jayshree Ullal [8:00]
Okay. Mine was 4K SRAMs.

Rene Haas [8:02]
Oh, my gosh.

Jayshree Ullal [8:03]
So what happened in those years is like, you know, first I got super excited and said, I have a zircom tester, I have a century tester. It's all my own to do things. And then obviously it was my path to immigration as well. You know, we talk about ​​​​​​​​H-1Bs, $100,000 a year right now. But, you know, back in those days, engineers were highly sought out. And we just couldn't get enough local people. So from there, I went to AMD, which is where I got my own train to networking.

Rene Haas [8:30]
Why did you leave Fairchild to AMD? I'm always curious, because I left TI after about three or four years, not really knowing all the opportunities that TI could have given me at the time. But what was your reason for leaving to go to AMD?

Jayshree Ullal [8:40]
I got laid off.

Rene Haas [8:41]
That's a good reason. Yeah.

Jayshree Ullal [8:42]
Yeah. So it's a very humiliating experience, which is why I would never do it to my own employees. But I should have seen it coming. They were like, you know, you're in Mountain View and your job is moving to Puyallup in Washington and we're going to move all the folks there. And I said, okay, well I'm not moving there. And they said, All right. And they absorbed that piece of data and laid me off. Right, which was the right thing for them to do for, you know, most of the things happened not because they're taking it personally out on you, but it was a cost cutting measure. But it certainly made me think about what next. And so I took a little bit of a break in India and came back and, you know, I had interviews with a number of the semi companies. It was clear that with my semi background, I’d be – my husband ended up in Intel and I in AMD, which made for great Christmas parties. When we went to see Jerry Sanders parties.

Rene Haas[9:30]
And was this late 80s, early 90s?

Jayshree Ullal [9:31]
Yeah, mid-to-late eighties.

Rene Haas [9:33]
And AMD, also product engineering or something different?

Jayshree Ullal [9:35]
Yeah. So from there. I actually did start out in product engineering, but I had an aha moment where, you know, one of my chips was failing and it had a packaging bonding problem, a Bit error rate. And I had to do a lot of, you know, schematic review and check the chips, check the bonds. And, you know, it was my first exposure to customers in a somewhat unfriendly setting because they were all angry that the chips weren’t working. And so did a lot of yield analysis and engineering analysis. And it got me thinking that there's another side of me, after having been five years in this field that I'd like to exercise. So I went into a field called Product Planning, and some of my best learnings came from defining the networking chips that we went on to build, including FDDI and Taxi, which was a parallel to serial converter to cut down the number of cables.

Rene Haas [10:29]
I remember those products.

Jayshree Ullal [10:30]
So great years where every time you wanted to design a product, you actually had to get in front of the CEO and have a justification like you do in front of a VC for a company, right, to get it funded. So part engineering and this is when I really got into applying my engineering into product planning.

Rene Haas [10:46]
So refresh my memory. AMD, I remember, was big into networking chips back in the day.

Jayshree Ullal [10:49]
They were, they were.

Rene Haas [10:51]
But were they attacking kind of a custom network or token ring? I'm trying to think now, but AMD back in the day –

Jayshree Ullal [10:55]
It was high speed token ring. It was called FDDI, 100 megabit. But we were also very involved in the early days in Ethernet, along with Xerox and Digital Deck. Again, now we've gone from mainframes to super minis, you know, we’re moving up in the stack here. And we even did some one megabit StarLAN, but my greatest memories were the 100 megabit FDDI backbone chips we designed.

Rene Haas [11:18]
Which were super high speed for back in the day.

Jayshree Ullal [11:19]
Really high speed. But they were designed, you'll get a kick out of this, on LSI gatarays that were running at, you know, 10,000 gates. You know, we never talked about nanometer, they were all in micrometer technology.

Rene Haas [11:31]
Why gatarays and not custom design?

Jayshree Ullal [11:33]
Probably because we were still feeling our way into – and the cost of it too. So at that time, the early designs, you know, didn't even have much of a custom design component to it. You just sort of designed it and threw it to them. And through that experience, I realized, oh my God, by the time you define a chip, design it, and get it out to the market, it can be 5 to 7 years. Markets can go by then. And so that's how I went into the system side. I said, you know, this is fun, but I want a little more instant gratification. And joined a company called Ungermann-Bass, which was actually at that time bigger than Cisco, if you can believe it. So I ran there. And they competed with 3 Com, right. And again somehow every company I joined ended up getting acquired. This one got acquired by Tandem Computers, and which then in turn got acquired by Compaq, which then in turn is HP. So somewhere there's some UB there, but built the first bridges and routers even before Cisco did. They were XNS routers, not IP routers, but we worked very closely with Cisco. Cisco was 5 million and my division in Ungermann-Bass was 50 million.

Rene Haas [12:37]
And this is now probably mid nineties –

Jayshree Ullal [12:39]
Yeah, early nineties, ‘91, ‘92, ‘93.

Rene Haas [12:42]
Yea, pre-internet.

Jayshree Ullal [12:43]
Yes, pre-internet. And then I had to do what everyone does. Did you do your startup too? Did you do a startup? Which one was that?

Rene Haas [12:50]
I did Tensilica.

Jayshree Ullal [12:51]
I know that one. So mine was Crescendo Communications.

Rene Haas [12:55]
You made a good choice.

Jayshree Ullal [12:56]
Well, it was a short-lived choice, I was only there two years before we got bought in ‘93 as Cisco's first acquisition.

Rene Haas [13:04]
How many people were at Crescendo?

Jayshree Ullal [13:06]
At the time, probably 50, 60 people.

Rene Haas [13:09]
So I remember when I joined Tensilica, I had always had the dream of joining a startup. It was the one thing I wanted in Silicon Valley. I knew nothing about startups.

Jayshree Ullal [13:17]
You don’t until you get into it.

Rene Haas [13:19]
But what were the things, looking back, that really surprised you about that experience?

Jayshree Ullal [13:24]
Well, something surprised me, but I had worked with them – on the Ungermann-Bass side, we had worked on an FDDI copper spec with them and we were trying to figure out what –

Rene Haas [13:32]
They were your customer.

Jayshree Ullal [13:33]
They were not my customer, we were working on standards together. We were working on whether it would be PAM4, or what the technology would be. So I had a chance to get to know the people. And to me in a startup, people are the greatest choice you can make because you know, people can shape any technology and get a product, but products, you know, can only be shaped by people, right? So in some ways, it wasn't a shock because I'm like, okay, I think I know what I'm getting into. But I couldn't have imagined the complexity of, ‘okay, you can get a product, but how do you get it to market and how do you get it to your customers?’ And that's where I came in. And certainly it was, you know, it was an exercise of one by one. We were fortunate, we won two important customers. And in fact, Microsoft's been a big part of my life, not only in Crescendo, but Cisco and of course, now Arista. They were one of our early customers, and Redmond had just started then. Their new buildings were just coming up. It was an FDDI backbone. We got a chance to work with them. I don't think I could have estimated the complexity of building a technology versus going to market with a technology. It was hard.

Rene Haas [14:38]
I mean, back then, a lot of these companies were being acquired, snapped up superfast by Bay Networks or Nortel or 3 Com. Was Crescendo, when they started it – did they view that ultimately this is going to be an acquisition or we're going to go IPO?

Jayshree Ullal [14:51]
Not at all. In fact, they were all very sad that day. You have to remember, back in ‘93, acquisitions were still not a fad and we got bought for 90 million, you know, it's pretty pittance if you think of it in today's age. But it gave us all a sense of accomplishment. And at that time, Cisco was less than a billion.

Rene Haas [15:09]
For Cisco, it’s a huge acquisition – in terms of strategy.

Jayshree Ullal [15:13]
Undoubtedly. Cisco was a company at roughly 800 million revenue. And I got a chance personally to go build their switching business and build and name the catalyst switches, which went on to become a $10 to $15 billion business.

Rene Haas [15:28]
That was the growth engine for Cisco.

Jayshree Ullal [15:30]
Absolutely. And I think part of it is, I think we all agonized about going there and we had signed agreements to stay there two years. But we ended up being there 15 years, two years at a time, right, and I think we contributed greatly. But Cisco also gave us a platform to contribute. It was both ways.

Rene Haas [15:45]
One thing that Cisco did – and as you said, you were one of the early ones, and then Cisco just went on a tear, whether it was Granite, etc, etc – did an amazingly good job of somehow integrating these startups into the company and culturally, looking back, I’m just wondering how did they work so well? What did they do?

Jayshree Ullal [16:02]
I think John Morgridge was the CEO then. John Chambers was the head of sales, and they both taught me different lessons. ​​​​​​​​John Morgridge was like, ‘look at your margins. They're in the fifties. When are you going to get them up to our number of seventies and eighties?’ I said, ‘All right, we've got to do some cost reductions and start building A6 and all the things you have to do to make your boards’. John Chambers, though, was, you know, a wealth and depth of knowledge in how to go to market. We didn't have much of a sales force, but he was instrumental in investing in a sales force. And he told them all, you know, ‘don't just sell routers, you got to sell switches, they are the future.’ So I think they both came at it and I thought I'd go nuts listening to two of them. One wants high margin, one wants to get to a billion. There's no way I'm going to last here more than two years. But I think the biggest credit I would give Cisco at that time is they identified the talent and people, nurtured them, and left them alone. And that was a huge part of –

Rene Haas [16:55]
How did the engineering cultures mix or did they need to, in other words –

Jayshree Ullal [16:58]
They didn’t need to.

Rene Haas [16:59]
You could be fairly self-sufficient.

Jayshree Ullal [17:01]
Here's a good example. So Cisco had iOS and–

Rene Haas [17:05]
And that’s old iOS, that’s not the iOS that most of our listeners would think of –

Jayshree Ullal [17:08]
No, not today's iOS. But the original iOS, I should say, back in the nineties. And when Crescendo came with Catalyst, we brought in something called CatOS, right? And so CatOS and iOS were, you know, kind of two headed monsters. Even today, when I talk to customers who see the Catalyst 6500 as the most successful product in the industry, they'll go, I like that product because it had iOS and somebody else would go, I like that product because it had CatOS. But we actually built a two headed monster because we just couldn't get it all in one. And practically speaking, if we had tried to get it all in one, it would have taken us five years. So it would have been a long time and you would have seen the market go by.

Rene Haas [17:43]
So you're running both operating systems?

Jayshree Ullal [17:44]
Yeah, so we're on a switching layer two basis, we ran the CatOS and then when we needed the layer three, we brought in the iOS as a routing blade into that. So like I said, it was a two headed monster. Many didn't like it for that. And then we tried to mix and match it, but it was the right call from a business point of view because we were able to basically go from 40 million in revenue in the first year we were acquired to 365, which I remember that number well because it was million a day for the year, to a billion the third year. And you know, obviously this is the wave of AI and that was the wave of the World Wide Web and internet.

Rene Haas[18:19]
Well, Crescendo then ended up being the switching business for Cisco.

Jayshree Ullal [18:22]
That's right. We ended up buying Kalpana and Granite and Grand Junction, because we built the high-end switches, but then we needed the portfolio.

Rene Haas [18:29]
So how did you, I hate to use this word, but assimilate the Granites and the Kalpanas of the world into the Cat-5000 family engineering-wise, again were they bolt-ons or –

Jayshree Ullal [18:40]
No, we tried to keep their strengths, each of them. I mean, you look at Grand Junction and Granite, they have fantastic strength in building chips and E6 and we wanted to keep it, but we also tried to enforce the portfolio. We’re like, ‘hey, you can’t go off and do another OS. Come and take ours.’ So there were some moments, but it was really good portfolio management and the teams really got along. We called it Workgroup Business Unit, WBU. And today, the reason I know Andy and his team so well is they were the basis for the Cat 4K and they worked to integrate where it made sense.

Rene Haas [19:11]
Cisco, again, is almost the boilerplate for the success story on M&A. Why were they so good at it? What was the blueprint that made Cisco so good at making that work? Because they were and are world-class in that.

Jayshree Ullal [19:22]
I think they had the success of a few good acquisitions that didn't just go well, but went very well. Which could make up for some of them that went less well. It's always like that. So, you know, if they did 40 or 50 acquisitions, maybe five and ten were very good and that's the ones we talk about. And the others were not as good, but they weren't bad either, right. So some of them were just homeruns. You know, or like in cricket, they were fours and sixes, rather than singles.

Rene Haas [19:48]
I just look at M&A, you know, on the top line you can look at it the way an investment banker looks at it and you see that the bar charts and the revenue and what's accretive. But to your point, relative to people– products come down to people and ultimately getting people to work together, particularly when you've got a small company culture into a large engine, that's hard.

Jayshree Ullal [20:08]
I didn't think I'd stay there, but part of the credit I have to give Cisco is they nurture the people and when they saw the success of the switching division, they let those people be leaders inside of Cisco. And, you know, I always tell people I was a misfit in Cisco because I didn't fit the executive ranks and talk the talk, walk the walk. But in some ways, I grew up in Cisco because they let me do it and make mistakes or make success. So I think a lot of credit goes to them for helping cultivate the entrepreneurial side.

Rene Haas [20:40]
I want to talk about Arista, obviously, and AI. But just one last thing on the Cisco thing. They did the spin out and spin in of Luca, Mario, and Prem. What made that work so well? All great guys.

Jayshree Ullal [20:52]
And great engineers. I think what started happening in Cisco, like it happens in many corporations, is the ability to innovate slows down no matter how good you are because you're a victim of existing products and things you're doing, right, so the only way to get more innovation was to cordon them off, if you will, and you could cordon them off in a different building, or you could cordon them off to spin ins. And so you have to also remember hiring talent was not trivial at that time. So the combination of needing talent, putting a focus, not letting them worry about the legacy business, the traditional business so they could go do this, I think made it very successful. The flip side of that, which I never did and I never was part of it, is because it did create a huge morale issue when they came back on the haves and the have nots. Right, because there are very many good engineers there. So I think it worked very well as long as they built products that were complementary to what Cisco's already doing. Then it was a natural. Nobody could object because they were adding accretive value.

Rene Haas [21:52]
So it's a win.

Jayshree Ullal [21:54]
Exactly, win-win.

Rene Haas [21:55]
So tell us about how 15 years at Cisco, you're doing amazing. You're running a giant business. What happens with Arista?

Jayshree Ullal [22:01]
Right - you know, I was in my mid-forties and there's this moment of, I guess, midlife crisis, if you want to call it that, where you kind of say, ‘oh, are you going to retire here or are you going to do something else?’ Right. And I felt like Cisco had become a very big company, but I also felt like I had contributed and done all I can. And while I wasn't sure exactly what I was going to do next, it wasn't clear it was Arista. I knew if I stayed on and got comfortable, I would never do anything different. So it was one of those, you know, forcing yourself to make a decision and think about it, where I knew if I had the gas in the tank to do one or two more, that now's the time to think about it. It was less about leaving Cisco and more about not knowing what I'd do next, but wanting to do something next.

Rene Haas [22:47]
Yeah, very similar to me.

Jayshree Ullal [22:49]
Yeah, I remember you telling me the same thing, right?

Rene Haas [22:51]
There was nothing wrong.

Jayshree Ullal [22:52]
Which is the best way, right?

Rene Haas [22:54]
Yeah. Wonderful company. And obviously what they've done is legendary. But for me, it was about scratching a different itch.

Jayshree Ullal [23:02]
Right. And in fact, I had a lot of people say, you've got a powerful position. Why are you leaving? The company is great. But I think, like you with NVIDIA, you get to a point where you say, ‘oh, that's true, but how can I make a difference again?’ And what I enjoyed at Cisco greatly was the impact and the contribution I could make to them. And is it possible to do it again? And I have to tell you, while I did it, there was an incredible fear of failure. I have to admit that. I was scared about that decision too, I had no idea what I was going to do. And everybody looks back and, you know, it looks like a rosy dream and journey come true. But it wasn't without its moments.

Rene Haas [23:37]
And the role you took was the CEO role at Arista?

Jayshree Ullal [23:40]
Yeah, so I took the summer off. I actually looked at a lot of different technologies. And security and, you know, energy and clean tech was very big at that point. So Vinod Khosla had a lot of companies I was looking at, but I ultimately concluded that I didn't have a Ph.D. on those topics. I had a Ph.D. in networking through work, not through a degree.

Rene Haas [24:04]
I love how you put that. That's pretty accurate.

Jayshree Ullal [24:05]
Yeah, except for the honorary degree I got for speaking at Santa Clara graduation. But the point was, if I wasn't going to leverage what I knew and build on it, and if I wanted to start again, then I would be making myself a whole lot more uncomfortable. And I don't know if I could contribute in the same fashion, right? So Andy and I have been friends back when he says we were kids. You know, he says I was a little girl. But I remind him, he was a little boy. I was in my twenties, he was in his early thirties, when I was AMD and he was at Sun. and he said, ‘Hey, I’ve funded this company and we're building the software stack.’ And they were just shaping what they wanted to do. They were thinking, should we build a switch? It was largely software. They weren't building any merchant silicon and hardware in any substantial way. And I said, Oh, switching, Andy, I just came out of that. Why would I do this again, right? But when I met the team, there was, again, back to the story you and I have talked about many times. It's the people. And I came away very impressed with the founders can do to build this amazing extensible operating system. And we didn't exactly know how it would shape. It was zero revenue, it was 30 engineers. But I remember Andy and I having this discussion saying it'd be great to go from 0 to 100 million, right? And I said, ‘yeah, right.’

Rene Haas [25:21]
So pre-revenue -

Jayshree Ullal [25:22]
Pre-revenue, zero.

Rene Haas [25:23]
About thirty people.

Jayshree Ullal [25:24]
30 engineers.

Rene Haas [25:25]
That's a very fun and scary time.

Jayshree Ullal [25:27]
Yes, it is. We were at the basement of a lawyer building that has since gone bankrupt. And we lived to tell the story.

Rene Haas [25:34]

And that was 2000-

Jayshree Ullal [25:36]
That was 2000..8. 2008. Right in the middle of the financial crisis and recession.

Rene Haas [25:40]
Incredibly brave.

Jayshree Ullal [25:42]
That's what everybody says. That’s what everybody says, yes.

Rene Haas [25:48]
And when did Arista go public?

Jayshree Ullal [25:47]
2014.

Rene Haas [25:48]
2014. So not long after.

Jayshree Ullal [25:49]
Six years. And it took six years because –

Rene Haas [25:51]
That’s very, very fast.

Jayshree Ullal [25:53]
In one way it's fast, but in another way, you know, we did take our time. We took four years to build a software system and another six years to productize and get revenue. We wanted to get substantial size. You know, be careful what you ask for. You know, you go public and then, as you know, you've got another whole constituent to please with shareholders. So we were north of 400 million when we went public with a market cap of 3 billion or so. But the best news of it all was, we had a product, you know, of high frequency trading, low latency. We had entered some of the world's largest cloud customers and defined and pioneered a leafspine architecture which until then was maybe done in HPC by Infiniband.

Rene Haas [26:38]
That’s huge.

Jayshree Ullal [26:39]
And I think the barrier to entry with our software was undeniable, you know. So I think getting all the trifecta of great innovation, great team and great customers was important.

Rene Haas [26:46]
When you took the job, thirty engineers, 2008, how far could you see? In other words, how big did you think it could be?

Jayshree Ullal [26:53]
It was one step at a time. I think we thought we could get from 0 to 100 million. I don't think we imagined a billion or 5 billion or 10 billion that I promised the street in 2026. Ever.

Rene Haas [27:06]
You never imagined that.

Jayshree Ullal [27:07]
No, no, I would be kidding. I was not clairvoyant enough to think that way. However, we did know we were attacking a large market, but we also knew, having been at Cisco, we had deep regard and respect for what they had accomplished over decades, not years. So we were realistic about really focusing on the niche markets.

Rene Haas [27:25]
So when you were doing all hands and people would ask the question about how big of a company could we be, how would you answer that question?

Jayshree Ullal [27:31]
Yeah, that was probably a bad answer on my part. But we would say, ‘yeah, you know, we're not competing with Cisco, we're a niche networking company. We could get to 100 million, but there's a path to a billion dollar valuation’. I remember Andy talking about the road to a billion and I’d say Andy, how can we say that when we have no valuation right now? So he was the bolder of the two. I was the conservative one. But between the two, I think what we could definitely tell the teams at all hands is this is a special company. It's got a neat culture, its engineers, you know, driving engineers, even our customers were engineers. It's prioritizing innovation over sales quotas or commissions or just selling the last thing we wanted to sell and that we were building something with a strong foundation and a long journey. So that naturally attracted a set of people and it didn't attract some people, for sure.

Rene Haas [28:23]
Percentage of the original thirty who were there when you got there?

Jayshree Ullal [28:26]
More than half.

Rene Haas [28:27]
That's incredible. Yeah, it's actually really incredible.

Jayshree Ullal [28:28]

Yeah, absolutely.

Rene Haas [28:30]

Yeah. And now, you know, inside our company, I tend not to like to talk about the numbers too much because to me, they're kind of outcomes. But when we think about the opportunity in front of us now with all things with AI, it does appear almost unbounded.

Jayshree Ullal [28:46]

Yeah. You know, there are days we all have been through the bubble where you go, is this a bubble? But I think not for a number of reasons that I'm sure you've thought through, too, which is there may be moments of highs and lows, but there is a genuine need to improve applications and software and therefore a genuine need to build an infrastructure to do that, right? So unlike the Internet bubble where you had a whole bunch of sea legs that kind of went, ‘build it and it'll come and we'll wait for it to come’. These are responsible companies that are building up this infrastructure that have financial depth, but also conviction to spend more, to do more, right. So I think it is unique times. It may not last forever, but I think it's – just like the Industrial revolution and the Internet revolution, I think the AI revolution is very real.

Rene Haas [29:28]
It’s a good analogy because the Internet revolution, you remember that, yes, it was a bubble in the sense of the spend versus demand. But you had companies like Global Crossing and things of that nature that just vanished off the face of the Earth.

Jayshree Ullal [29:39]
Oh, my God, they did, yeah.

Rene Haas [29:40]
And then, the other big difference is, back in 2000, the most valuable companies on the planet were not Microsoft, Meta, NVIDIA, Amazon. We're now in a world where this growth is being fueled by, you know, the largest companies on the planet.

Jayshree Ullal [29:55]
Exactly.

Rene Haas [29:56]
There is a part of me, though, that looks at it and says at some point, does gravity take over and can we really make use of 100 gigawatts, a terawatt? But to your point, the dimension of problems that are unsolved that could be solved by artificial intelligence definitely appears to be unlimited. So I'm very much more on the bull side than the bear side. But it's going to be a fascinating thing to see where it goes.

Jayshree Ullal [30:18]
I think, you know, you could argue that the training models require all of these multi gigawatt, but you could also argue that you can have a distributed inference set of models, too, that require multiples of megawatt. Maybe it aggregates into a gigawatt, but whichever way you get there, going back to our analogy is we went from mainframe to super mini to client server to the PC to the phone, right? So it may come in a mini me mode or it may come in a more macro mode like it does right now. But it's going – it's here.

Rene Haas[30:49]
We're going to be serving tokens all day long.

Jayshree Ullal[30:51]
Small, medium or large. Absolutely.

Rene Haas[30:53]
So it's obviously been a huge success with Arista, long career at Cisco, AMD, Fairchild.

Jayshree Ullal [30:58]
And congratulations to you on Arm. It’s great to watch that as well.

Rene Haas [31:00]
Thank you so much. Who are your mentors that you've had in your career? A, any you'd like to name and B, what did they teach you?

Jayshree Ullal [31:08]
Well, I think along the way I've had lots of mentors. Let’s start with my younger age. My dad and mom were heavy influencers. They raised two daughters and they raised them to do whatever they could do and pushed us to the limits, right, my dad was a physicist himself. So, you know, he believed in tough love and he gave a lot of tough love. But definitely, I would say they were my early mentors. My family, my husband and everybody, huge supporters. You can't do what we do without that support. I still remember my husband encouraging me when I was in pure engineering and saying, ‘You can do it, you can go to the other side’. And I don't consider myself a trained business person. I don't have an MBA, but, you know, live and learn, as they say, right, so I think that's a huge part of – the family support is still important as well. You know, along the way, I've had some great bosses. Without naming them, some of them were in AMD. You may know them, do you know John East? John East was a wonderful human being, a great mentor and people there who trained me – everything from, okay you’re taking your first flight to Boston. ‘Let me show you on the map how you land and take the Sumner Tunnel and deal with those aggressive drivers there.’ Everything from how to travel to the first luggable I had, I think it was a compaq luggable top or whatever. And to my professors, I owe a lot to my physics teachers, my professors in undergrad who came to my going away when I left Cisco. And it's so nice that they remember me as a studious nerd, but for some reason they also remember me as having potential, which is nice, right? At that time, you're so not confident of yourself. And then obviously through my career, you know, I would not call them just – I would call them colleagues, mentors, friends, because after a while it isn't who you work for, it's the circle you form.

Rene Haas[32:52]
It is the ecosystem. Absolutely.

Jayshree Ullal[32:54]
It is the ecosystem. And I'm proud to call you a member of that ecosystem.

Rene Haas [32:57]
Ditto. And as we chatted before, it's amazing the similarities of our backgrounds, how we all got here – but successful CEO, successful mother, grandmother. Jayshree, thank you so much for joining us.

Jayshree Ullal [33:09]
Thank you for having me, Rene. It's an absolute pleasure.

Rene Haas [33:17]
Thanks for listening to this month’s episode of Tech Unheard. We’ll be back next month for another look behind the boardroom door. To be sure you don’t miss new episodes, follow Tech Unheard wherever you get your podcasts. Until then, Tech Unheard is a custom podcast series from Arm and National Public Media. And I’m Arm CEO, Rene Haas. Thanks for listening to Tech Unheard.

Rene Haas [0:08]
Welcome to Tech Unheard, the podcast that takes you behind the scenes of the most exciting developments in technology. I’m Rene Haas, your host and CEO of Arm.
Today, we have a very special episode for you. I’m in Bangalore, India, joined by Ashwini Vaishnaw, the Indian government’s Minister for Railways, Information & Broadcasting and Electronics & Information Technology. Under his leadership, India’s electronics manufacturing and exports have grown notably and the country’s first domestically produced semiconductor chip was released this fall.
Minister, thanks so much for joining me.

Ashwini Vaishnaw [0:45]
Thank you.

Rene Haas [0:46]
Pleasure to have you. And you joined us today for this grand opening, which was also pretty amazing. Have you done many of these grand openings of offices across India recently?

Ashwini Vaishnaw [0:56]
Not many, but Arm is special, so I had to be here. And good thing is that you're doing two nanometer design here. So that makes it a very special occasion for me, so that's why I came here.

Rene Haas [1:10]
Yeah. So there's a lot of things I want to talk to you about. There’s some amazing things going on with Indian economic growth, the things that are going on with semiconductor initiatives in India, which has a long and rich history. But maybe I want to start, Minister, with your background, and your background is unique. You've been in the technology space, you've been in the private sector, but now you're a high-ranking official in the Indian government. So, tell us a little bit about your journey and how you got to where you are.

Ashwini Vaishnaw [1:34]
It's an interesting journey. And see, I started as an engineer. I studied electronics and communication, and during college itself, I got hooked to semiconductors. And in fact, we somehow got one chip in our college and invented – used that to invent a network printer, way back, that’s a long, long time.

Rene Haas [1:55]
And what were you programming the 8085 in?

Ashwini Vaishnaw [1:57]
That was hex.

Rene Haas [1:59]
The hex assembly? My goodness.

Ashwini Vaishnaw [2:02]
It was very, very difficult and it required so much focus, I remember.

Rene Haas [2:05]
You probably still remember some of the commands, I would imagine. I still remember some from 8080. So, you studied engineering in university…

Ashwini Vaishnaw [2:12]
Correct. Did my MTech at IIT Kanpur and during MTech, again was very fortunate to work on transformer for text-to-sound. Actually, I designed a card which was manufactured by Intel and then I had to write the entire database management system, because there was no DBMS which would take sound as a data element. So, it was very interesting and very kind of – at that point of time, probably a very unique project. Then I got into government, worked as an IAS officer, which is a challenging task in our country, managed.

Rene Haas [2:45]
And what is that – for our listeners who may not know, what is an IAS officer?

Ashwini Vaishnaw [2:51]
So I’m sure people in India would understand. Outside India, one example I can give is like managing a complete district in India, maybe having like 1.4 million people, kind of thing. Having one large city or two large cities, and many small villages. So managing that, practically every aspect of governance comes to you, so that's very interesting. Then I worked in the Prime Minister's office, and then I went to Wharton and did my MBA, studied finance there, and came back and had a good entrepreneurial and corporate journey. And our Prime Minister gave me this very big opportunity in 2019. He got me into the parliament, and gave this big responsibility.

Rene Haas [3:37]
So you were designing text-to-sound, you were working on a network printer. You got your degree, your masters, yet you moved into government. What drove that choice for you at the time?

Ashwini Vaishnaw [3:51]
A lot of things happen in our lives which are driven by the context and the society of that period. A large part of it was driven by my family. My father wished me into civil services, but I was more inclined towards technology.

Rene Haas [4:06]
And were there aspects of your engineering background that when you got into civil service that helped you? Engineers are known to be very, very logical folks. Were there aspects of it when you look back that served you well?

Ashwini Vaishnaw [4:21]
It served me very well, because engineering gives you that logical way of thinking, that problem-solving mindset comes when you do engineering training and that really helped me. In fact, in one example I can give is the ‘99 super cyclone which came on the east coast of India. At that time, I was serving in a district called Balasore. I practically went into the computers of NASA. And got a lot of good data and collected that together and could predict the path of the cyclone. So that really helped.

Rene Haas [4:52]
Yeah. The thing that I'm just so impressed with is - and we’ll talk about your role now - is that having that engineering training, both in terms of your education and doing work in the private sector, has you rather uniquely positioned to understand the needs of the marketplace.

Ashwini Vaishnaw [5:07]
It helps me understand it. It also helps me engage well with the technical world. One can understand the challenges that we would face and one can foresee the entire spectrum of the issues that you’d face.

Rene Haas [5:20]
So, tell the world that's listening here, your role now, what you do inside of Parliament for the Ministry. You've got - your title’s rather unique in terms of the areas you look after, which are quite large and broad, but tell folks a bit about what you're doing.

Ashwini Vaishnaw [5:35]
So currently I've got three responsibilities. One is electronics and IT, where we basically are focused on manufacturing, on developing solutions for digital public infrastructure, and on creating the semiconductor ecosystem in our country – that is one, and the AI piece of it, AI mission. Second is railways where we have a very large network, which is more than 150 years old. A large population uses railways, so modernizing that entire network, getting it up to speed towards today’s technologies. And third is information and broadcasting.

Rene Haas [6:13]
So let's talk about semiconductors. I started my career in the middle 1980s and at that time, Japan was overtaking the United States and having a national policy on semiconductors was extremely important. And the U.S. started a consortium back in the day to address that. And then there was a period, I think, when the Internet boomed, that semiconductors weren’t seen as so interesting anymore – national policy waned and there wasn't as much attention given to it. Now the world has changed for a myriad of things. When you think about it in your role, what are the things that come to mind in terms of how you think about semiconductors, both in terms of the importance to India as a country, but also how you think about a national policy?

Ashwini Vaishnaw [6:57]
So if you look at it, any country which wants to be a significant player in the world must have good command over certain technologies. Semiconductor is one of them, quantum is another. getting the capabilities to work on those technologies. Telecom is another. So it's very important that we have significant capabilities in these technologies. That's the way we look at semiconductor as a major technology in our entire constellation of spectrum of technologies that we are forming. We have very strong design capabilities in our country. Close to one-fifth of the design talent is here in India, globally. So, every large global semiconductor company has a good presence here.

Rene Haas [7:47]
Some of our best people working on the most advanced products for Arm are here.

Ashwini Vaishnaw [7:51]
Correct. And really advanced chips like two nanometer, three nanometer, some of the advanced nodes are being designed here. Based on this design capability, we want to now get into the manufacturing. So fab, ATMP, the capital equipment, the materials, and the finished product, which is electronics, right? This entire spectrum we are working on. So, in the case of semiconductors, we definitely want to have a good design ecosystem, further develop startups which can be – basically, which become product companies. And we want to do fabrication and manufacture of chips. We want to do ATMP. We already have ten units under construction at this point of time as we speak. Two of them are fabs and eight are ATMP units. We want to also manufacture the equipment which goes into manufacturing chips.

Rene Haas [8:43]
And, sorry to interrupt you, Ashwini, but you started in 2019 in this role. When did you start to put in place this thinking about, we need a national policy around semiconductors?

Ashwini Vaishnaw [8:53]
So our semiconductor mission was inaugurated on the first of January 2022.

Rene Haas [8:59]
Okay.

Ashwini Vaishnaw [9:00]
2021 is when our Prime Minister approved the semiconductor mission. And then first January 2022, we started the mission. And in three and a half years, we have made significant progress.

Rene Haas [9:11]
And here, relative to – here being in India – how does the government get involved relative to the private sector, the university system, to distill that mission into these partners to help build the ecosystem. How does that happen here?

Ashwini Vaishnaw [9:27]
So the mission is run by professionals, technical people, people who understand semiconductors very well. They are the ones who are running the entire mission, it’s run in a very professional and transparent manner. Our interaction is with multiple stakeholders, with the chemical, the gas manufacturers, the equipment manufacturers, the designers. In terms of universities, we have tie-ups – within the country, we are developing a very large talent pool, 278 universities we are working with.

Rene Haas [9:59]
So, 278 universities in India. And how does the government organize that effort with the universities, how does that take place?

Ashwini Vaishnaw [10:07]
So, we have a body called AICTE – All India Council for Technical Education. That's the coordinating body. Under the semiconductor mission, we have provided free-of-cost access to the world’s best EDA tools to the students studying in these universities. Basically, we would like our students to come out of the college well-prepared for the industry and also having that edge over other parts of the world.

Rene Haas [10:33]
Yeah. That's fantastic. And the vision is not just designing chips, but your vision is to go very, very broad with this.

Ashwini Vaishnaw [10:42]
Our Prime Minister gave us the mandate to develop the entire ecosystem. So, we want to manufacture chips, we want to manufacture the equipment which goes into manufacturing chips, and we want to manufacture the materials which go into manufacturing chips. So, the complete ecosystem we are working on.

Rene Haas [10:57]
Those are bold ambitions, and I think they're the right ones. When you think about the headwinds, things that make that challenging, what comes to mind?

Ashwini Vaishnaw [11:05]
When you start any industry, there certainly would be many challenges. And when we started this journey, I remember my first interactions with the semiconductor professionals on the west coast of the U.S. – I met about 45 top CEOs and CTOs – and each time my pitch used to be: “We are starting a new industry and I don’t know anything. So please help us understand what this industry is and what we should be doing.” And everybody guided us very well and I must thank each one of those 45 CEOs who helped us think through the challenges. And we meticulously looked at each and every challenge, and: “Okay, this is a problem about power supply, let's find out how do we mitigate this.” “This is a problem about a particular very high ultra-pure chemical, let's find out how to mitigate that risk.” And methodically we have been working.

Rene Haas [11:58]
So, semiconductor investment for the type that you're talking about is not a one-year thing. It's not a two-year thing. It's a decades kind of thing. How do you ensure that the momentum lasts decades, given the fact that you're in a country here in India that has elections, government leadership could change. How do you think you maintain that sort of “stick-to-it-iveness”, if you will, relative to continuing that journey?

Ashwini Vaishnaw [12:27]
By having a very large ecosystem and a very large number of stakeholders. One good thing about our country is once we take a direction, the momentum once it builds up, it remains maintained for a very long period of time. And when you have a very large ecosystem where everybody is partner in the program, then it helps in creating that sustainability.

Rene Haas [12:49]
Are there parts of the ecosystem that you think need more investment relative to the semiconductor vision?

Ashwini Vaishnaw [12:56]
I think the chemical and gas part will require a lot more attention. Because the ultra-pure chemicals which are required in manufacturing chips will require a lot more capabilities than we have.

Rene Haas [13:09]
Yeah, I think from a talent standpoint, certainly speaking for Arm, we're – not only we have a huge presence here, but some of our best people are here. I think that there's such a rich talent base of engineers here and also entrepreneurs who are here and who have moved to other parts of the world and come back – I feel like talent is not an issue, but maybe I'm not thinking about it the right way. Do you think…

Ashwini Vaishnaw [13:34]
There are two parts to this question. Basically, the talent which designs and the talent which works in the clean rooms. So, we are working on both parts of it. The design part is – that pool is already available and there's a very large number of those. But the clean room, the people who currently work in fab, that’s where we are currently focused on, so that we get the right people.

Rene Haas [13:56]
Now semiconducting, I want to come back to it, but I want to talk a little bit about India's economic growth and where that's going. It's obvious to anyone who comes here, and you and I were just remarking, looking outside this beautiful new office here, you could be anywhere in the world when you looked out at these gleaming buildings. It feels like this nation is on the cusp of some hyper-growth. Tell us a little bit about how you think about the growth engine of India.

Ashwini Vaishnaw [14:19]
So, it's a very well thought through strategy that we are executing. Our Prime Minister has given us a very clear way forward. It's basically built on four pillars. The pillar number one isinvestment in physical infrastructure as well as investment in digital and soc al infrastructure. Pillar two is about focus on manufacturing and innovation. Pillar three is inclusive growth and pillar four is simplification of so many processes and legal systems that we have. All the four pillars we are working in a very methodical way. We used to do public investment of about $30, 35 billion U.S. dollars ten years ago. Now we do about $140 billion dollars, and –

Rene Haas [15:02]
And so that’s capital investment in –

Ashwini Vaishnaw [15:05]
In railways, highways..

Rene Haas [15:08]
$140 billion dollars.

Ashwini Vaishnaw [15:09]
– Power, in building new universities, improving the suburban transportation, so many other things. The second part is focus on manufacturing and innovation. So starting from just about 300, 400 startups ten years ago, today we have more than 130,000 startups in our country and more than a hundred unicorns. We are today the third largest startup ecosystem. And Make in India, which is part of the second pillar, we are focused a lot on manufacturing practically everything. So electronics manufacturing, we have grown six times. Electronics exports, we have grown eight times. Telecom equipment, we always used to import ever since the 1950s, and now we are a major exporter of telecom equipment. Defense equipment, we used to import, now we’ve become an exporter. So, all those things are very much well-thought through part of the second pillar of manufacturing and innovation. Third is inclusive growth. We are a very diverse country, culturally, linguistically, economic growth level, all those things. So, we are very, very keen on making sure that the people at the bottom of the pyramid, they get significantly good opportunities to come up, and the economy, the entire society and the country should rise up as a good harmonious way. So many programs, for example, opening 540 million new bank accounts. And so that's huge, right? And any sector, like 130 million households getting tap water connection. 40 million houses built for the poorer sections of the society. It’s a very, very wide program, which we have taken up. Fourth is simplification, we have removed about 1,500 laws from the statute book and about 35,000 compliances from the system, and still that's a work in progress. We are still working towards simplifying the system. So, it's a very well thought through strategy. I can say with a high level of confidence that we'll continue to grow in the band of 6-8% over the next few years and then move higher in that band with very moderate inflation.

Rene Haas [17:25]
That's just amazing. And removing all those processes is somewhat music to my ears. I want to talk about AI for a bit – to what level does AI accelerate that? Both in terms of the technology, and then I'm very curious about how you in your role view using AI to enable some of the changes you're talking about.

Ashwini Vaishnaw [17:47]
India is an early adopter of AI. We are using AI in practically every sector of the economy today. We also have set up an AI mission. That mission basically does three major things. First is provide common compute facilities to our youngsters, to our startups and researchers.

Rene Haas [18:08]
So wait, what does that mean, common compute?

Ashwini Vaishnaw [18:11]
So if you look at the world, access to technology is limited to a few. So for example, in the case of semiconductors, we provided EDA tools to a very large population of students. Similarly, we are providing GPUs, 38,000 of them, to the students, the researchers, startups. So that common compute facility we created using government funds. And we are developing our LLMs, five teams are working on it. Then we have set up an AI Safety Institute, which is working in a very interesting network approach, which is developing a techno-legal solution to AI safety. We are developing many AI-based applications for our agriculture, education, healthcare, climate-related applications. So, it’s a very comprehensive program.

Rene Haas [19:00]
On the techno-legal approach towards AI, what does that mean exactly in terms of…techno-legal?

Ashwini Vaishnaw [19:08]
So we think that something as complex as AI cannot be dealt with simply by passing a law, by legislating something. It has to be solved in a more practical way. The practical way, we think, is to develop technical solutions and complement that with the legal structure. That’s the way we are thinking and – so for example, detecting a deep-fake. How do you detect a deep-fake? So, we have IIT Jodhpur working on a technical solution to detect deep-fake with a very high level of certainty. So that's an approach. And we have taken many institutions as a part of this network, rather than some of the countries or some other geographies without taking any names, I would say that they just want to legislate something and believe that the problem will be solved. The problem has to be solved in a more practical way.

Rene Haas [20:02]
Can it be solved through a sovereign cloud where you have control of the models, controls of the weights? Or is your view that the world is flat when it comes to this and these AI models cross national boundaries? And this is a global coordination that’s required around AI?

Ashwini Vaishnaw [20:21]
Absolutely, it's indeed a global coordination that we will require, because the digital world doesn’t have physical boundaries. So I think the good part of new technology will be harnessed by our corporations, by our people, by our startups. But the harmful things will have to be globally coordinated and to me, it appears that over the period of the next few years, the world will come to a level of understanding where people will sit together and look at solving this safety part in a more coordinated way.

Rene Haas [20:55]
Are you an “AI for good” – are you a bear or a bull when it comes to the benefits of AI?

Ashwini Vaishnaw [21:02]
I believe that technology does bring a lot of good things to the society, it can help us solve population-scale problems. For example, one of the applications we have is aimed at detecting tuberculosis. And it’s very effective.

Rene Haas [21:16]
Absolutely.

Ashwini Vaishnaw [21:17]
Another one that we are working on is improving the productivity of our agriculture. And the results are phenomenal. So, I'm very practical, very rational on this topic. I would say that, let’s harness the benefits and let’s keep our society safe from the harms.

Rene Haas [21:32]
Are you surprised at how fast it has moved in terms of, we had our ChatGPT moment a couple years ago and the advancement relative to whether it's video, whether it’s voice, queries… Are you surprised at how fast it’s moved?

Ashwini Vaishnaw [21:50]
I think the trend is very clear. If you look at a few decades ago, the technology cycle used to be a few years. Then it shrunk to a couple of years, three years. Then it shrunk to like one year you have a totally new technology. So, the technology cycles have really, really shrunk in the past, so one can really anticipate that the growth which is seen today will continue.

Rene Haas [22:13]
A lot of talk about “it's moving so fast and then there’s going to be a wave of jobs that are just going to get eliminated”. And then there are folks who say with every technological advancement, there's always jobs that go away, but there are more jobs that are created as a result. And then folks say, “well, no, this time it’s different because AI can think, AI can do the kind of jobs that certain trained white-collar workers could have done”. Is it something at the government level here in India, it’s talked about and thought about, I’m wondering?

Ashwini Vaishnaw [22:43]
We definitely are concerned about the change which is happening in the economy because of AI and the potential disruptions in employment. Yes, that's a major point. And we think that the solution to that is to upskill our people very rapidly. We think that, for example, the large number of students that we are training in 5G technologies and the large number of people that we are training in semiconductors, that same scale we must train our people in AI, both in the use, as well as in development. In fact, we have taken up a program for a million technical persons to be trained in AI-related technologies.

Rene Haas [23:25]
And what would be examples of those AI?

Ashwini Vaishnaw [23:29]
Multiple things. At a very basic level, making sure that the data annotation capabilities are there. At a mid-level, making sure that the applications can be created in India, for the world. And at a bigger level, on the research side.

Rene Haas [23:45]
Yeah. At Arm, and I think really across semiconductor companies, we're seeing productivity gains from AI, there's no doubt about it. But still finding very skilled engineers to do complex work around chip design, particularly some of the most complex chips in the world, we still see a need for a lot of engineers, as evidenced by all the growth here. Do you have a viewpoint on that, when you think about the semiconductor world, is AI – do jobs start going away in semis with AI?

Ashwini Vaishnaw [24:14]
I think the change will be there, I think the change will bring many good aspects, and there will be some parts of it which will require a lot of adjustment. What’s your view on this? Because you see the entire world from your perch.

Rene Haas [24:27]
Well, you know, I view AI as sort of this Star Trek thing that I did not think in my lifetime I would be able to work on it, from the perspective of, “can machines think and solve problems?” I think so long as ideas can be created to create new things, then I'm not so worried about where jobs go. It gets interesting though, when the AI gets smart enough to figure out, “here's the next thing you should do relative to developing a product or solving a problem.” Then I think it gets interesting. I am very much “AI for good,” as you are as well. I think healthcare is the killer app. When you look at the amount of time it takes to do a drug trial or genomic research, having that accelerated to an incredible level is pretty, pretty amazing. What the world looks like in 50 years? That's a great question. I would have thought that would have been a question years ago, we would have said, “well, it’ll kind of look the way it does today.” I think that time has shrunk. I think in 10 to 20 years, it’s going to be a very fascinating world. But chips will be at the heart of it, there's no doubt about it. So, what you guys are doing here in terms of your policy is the right one.

Ashwini Vaishnaw [25:41]
What’s the change in the semiconductor industry that you think – what’s the trend that you see where a country like us where we are starting a new industry, we should be thinking about?

Rene Haas [25:53]
Yeah, so in India, I think, there's so much potential to create the world’s next great geography for everything you’ve talked about, because you have access to a gigantic talent pool, you have access to resources, even rare earth-type minerals you have huge access to. So there's no reason over time, India cannot become a great dominant player in terms of the whole vertical infrastructure, which is why I find what your policies are behind to be so exciting, and it’s so fascinating, you’ve gone off and had these discussions with all the different CEOs, you said, on the West Coast. To be sitting here with a minister who can talk about Synopsys and Cadence EDA tools and understand where those are, that’s a huge jump because when it starts at the top, relative to an understanding of the details, is a very significant thing. So, I'm very excited by it. And when I come here to our office in Bangalore and I see the energy level and the passion of people to create and invent, like you said, there’s an entrepreneurial spirit here that’s so rich, I think there's no ceiling for what India could do in this space. So, I think it's just amazing. Ashwini, thank you so much. It was wonderful chatting with you and thank you again for joining us today on Tech Unheard and our grand opening in our new office.

Ashwini Vaishnaw [27:14]
Thank you, Rene. Thank you so much for inviting me.

Rene Haas [27:15]
My pleasure.

Thanks for listening to this month's episode of Tech Unheard. We'll be back next month for another look behind the boardroom door. To be sure you don't miss new episodes, follow Tech Unheard wherever you get your podcasts. Until then, Tech Unheard is a custom podcast series from Arm and National Public Media. And I'm Arm CEO, Rene Haas. Thanks for listening to Tech Unheard.

Rene Haas (0:00)
Welcome to Tech Unheard, the podcast that takes you behind the scenes of the most exciting developments in technology. I’m Rene Haas, your host and CEO of Arm. Today, our podcast sounds a little bit different than usual. Carnegie Mellon University kindly hosted me for a conversation with Farnam Jahanian, the university’s president as part of his President Lecture Series. Farnam Jahanian came to Carnegie Mellon from a background in computer science and engineering, joining the university over a decade ago as the vice president for research. Before his tenure at CMU, Farnam spent time at IBM’s research center and Arbor Networks studying Internet growth and stability. He also led the National Science Foundation’s Directorate for Computer and Information Science and Engineering, home to many programs focused on building cyber infrastructure and a computing and information technology workforce. Farnam and I had an excellent conversation and I can’t wait for you to hear it. We actually recorded on the CMU campus in front of a live student audience.

Farnam Jahanian (1:09)
Well, Rene, first of all, thank you for joining us. Thanks for being here.

RH (1:13)
My pleasure. No, it's great.

FJ (1:15)
Really happy to have you here. Can you talk a little bit about your professional journey, in particular, you had an incredible career being part of a couple of startups. You moved to NVIDIA, and of course, you've had an amazing run also at Arm in various capacities, including serving as CEO. So tell us a little bit about your professional journey, and also, what have you learned during that that's influenced you, particularly your approach to leadership as CEO of Arm.

RH (1:42)
One of my favorite speeches is if you've ever seen a Steve Jobs commencement speech at Stanford, and he's got this classic line about connecting the dots, and you can only connect the dots when you look backwards at your career. I was very fortunate with my first job out of school was at TI in Houston in semiconductors. And back then, TI was the number one company in semiconductors and I was very intrigued in, when I was in university, around computer engineering and computer design, but also felt semiconductors seem like they're the base technology that makes all of this go. So I chose semiconductors, which was very fortunate in hindsight, because now, particularly when you look at what's going on with the CHIPS Act and tariffs and semiconductors now are front and center to everything. So, I think starting there was incredibly fortunate. I can absolutely tell you, in 1984 it was not something I thought was going to be a super strategic decision. Houston was also very warm compared to where I went to university, but I will tell you where I failed the IQ test. I had never been anywhere south of Cincinnati, and I remember I took a job trip to Houston in February, it was 72 degrees, no humidity, and if you know what Houston's like, you're gonna see where this is going. And I came back and I told my parents, I said, “Houston's like California. The weather is no humidity, it's blue skies. I'm going to Texas.” And there's no internet, so there's no ChatGPT, there's no fact checking on how stupid the thing that was to say. But I remember getting down there with my mom, and in June of ‘84 and it was 98 degrees, and I thought, “Oh, my God, what have I done?” But that was a great choice in hindsight, because big company, which meant lots of opportunity. I got to work in fabs, I got to work in design, I got to work in product engineering, I got to work with customers. And one piece of advice that I think I would give folks, or what I learned is that experimenting with something new and getting out of your comfort zone and just trying something because you think it's curious, that benefited me. And one of the things that I put a very, very high premium on in terms of people development and people that we hire is curiosity. Because you can teach a lot of things in terms of skills, you can teach a lot of things in terms of how to get things done, but curiosity is that pilot light that keeps you going, and if you’ve got a natural curiosity for doing things, you're gonna go pretty far.

FJ (04:02)
So how did you end up at NVIDIA? And how was that experience?

RH (04:07)
Yeah, so I had now been doing seven years of startups, none that had gone anywhere. So I I moved to Silicon Valley in the mid 1990s. I wanted to do a startup - didn't really know what a startup even meant, to be quite frank, I just thought it was an interesting thing to go off and try. And I did a couple of them that subsequently had done okay, but in the time that I was doing it, it was a very difficult time. I joined right before the dot-com boom.

FJ (04:31)
So this is in the late ‘90 or early

RH (04:33)
Late 1990s, around your Arbor Networks time frame, right? So late 1990s I joined Tensilica, which was a company doing, of all things, semiconductor, microprocessor IP. Again, back on how the dots connect, that's a company I'm leading now. What we were doing was a processor generator, where you could generate a custom CPU, you could add custom instructions. It was a mix of replacement of RTL and code, and the internet's taking off. So as a result, there's tons of companies doing custom chips for packet forwarding, all the things relative to security for the internet. We hit the dot-com spike – doom – and as a result, we had to lay off 40% of our people, we had to cut back revenue. And then I had a chance to do another startup, again, doing 10 Gigabit Ethernet, which, back then sounds like, ‘oh, that's a no brainer’, but the world was not ready to put down infrastructure for 10 Gigabit Ethernet, let alone one gigabit, and we were running out of money. And then, you know, at the time, I thought, “oh my gosh, I've got two small kids. I don't really have a medical plan. I should go join a real company that's got actually a 401k and can pay its employees.” So I got a call out of the random from a recruiter for NVIDIA and I first thought, you know, my background at that point in time had been around chip design, product engineering, microprocessors. I couldn't spell GPU if you gave me the G and the P. And I met with Jensen – NVIDIA was not a very old company at that time, probably 10, 12 years, and by the way, was doing well, but not – nothing close to what they are now. And I remember asking Jensen, I said, “Look, I'll be very honest with you. I can do a lot of things, but I don't know anything about GPUs.” And he looked at me, I remember, I can recollect very well. He said, “We invented the GPU. If you think I need you to help me with a GPU, you must be kidding.” That was my first learning of Jensen, and it holds true. But yeah, I landed there in 2006 and it was a fantastic experience.

FJ (06:28)
And after a few years at NVIDIA, I think seven years or so, you were recruited to go to Arm.

RH (06:34)
Yeah, so this is 12 years ago.

FJ (08:26)
Absolutely, it's gone sort of full circle. Now you have a great partnership with NVIDIA.

RH (08:30)
We have a great partnership with NVIDIA. And in fact, one of the big benefits that we got with NVIDIA was all of their accelerators were connecting to an external processor, x86, and for a myriad of reasons, performance, power, efficiency, flexibility, they went to an integrated chip approach. The first one was Grace Hopper, Grace being the Arm CPU, and Hopper being the NVIDIA GPU. And then now with Grace Blackwell, which is all fully integrated. So they're doing things with Arm that they really couldn't do with anyone else. So it's great for us, it's great for them, and they're a fantastic partner.

FJ (09:06)
By the way, for those of you who may not be familiar, go look up Grace Hopper, and you'll see something about the history of computer science and technology.

RH (09:14)
Absolutely.

FJ (09:16)
You're at CMU, so we’ve got to talk about AI. Throughout the history of AI, of course, we've seen periods where AI has gone through a rise and then what people of our generation would refer to as winters of AI. And there have been multiple cycles. Now, over the last several years, of course, with generative AI and a new generation of AI-based systems, it seems different. It seems very different from the AI of the 1970s and 1980s and even the 2000s. What's different about this latest generation of AI-based systems, and what has surprised you about AI today and where it's going potentially?

RH (10:01)
Oh, boy, that's a fantastic question. This has sort of been the Holy Grail, right, in terms of, can computers think? And then you get into a definition of, “well, what does think mean, right? Is thinking passing the bar exam where there's a known answer, or is thinking invention, creation, developing something that no one has ever developed before?” So we've been in and out of these winters. And I think back to the NVIDIA example. I think you had the perfect intersection of the huge amount of work done, maybe it's the AlexNet moment, the work that Demis and the DeepMind guys had done, combined with a processing tool that is pretty well suited for this kind of problem. Because I can assure you that when we were working on CUDA, thinking about deep neural nets in the early, early days wasn't one of the application areas. But, you know, look at the early days that work was being done – they were using NVIDIA gaming cards. They weren't actually even using anything that was purpose built. So I think it is the intersection, as many times happen with technology, where a lot of work is taking place on the software side, and then a lot of research is being done in areas like Carnegie Mellon and you had an intersection where now the tools are there.

FJ (11:15)
That's right.

RH (11:16)
And that's where we're at. Intellectually, I always felt like this was going to happen. We would see this kind of work. I just didn't think I would be able to work on it in my lifetime. I didn't think I would be able to be experiencing it. But now it's moving at such a remarkable pace, it's going to be fascinating to see where it really goes and how it really goes. On one hand, I think there's a little bit of sky's the limit in Stargate and the things we're doing are around that. Then there's another part that says, “well, I don't know yet. Are we stochastic parrots with these LLMs, or are they really, really thinking?”

FJ (11:48)
You know, I'm so glad you mentioned the computational resources that were needed to achieve what we're achieving today. You're right that advances in algorithms and LLMs - undeniable in terms of how far we've come over the last couple of decades - but without advances in computational resources and being able to do this, we would have still essentially been looking at AI and not seeing the kind of transformation that we're seeing. So, a follow up to that question, how is Arm meeting this moment? You talked about the challenges, you talked about the speed at which technology is moving, how is Arm meeting this moment?

RH (12:31)
So we're putting a lot of energy, no pun intended, into this. So today we are involved with NVIDIA's Grace Blackwell, and most of the work that we do in that is, the GPU handles a lot of the heavy workloads, you obviously need the CPU to do a lot of offload management of the system, you know, etc, etc. But at the highest level, the good news for Arm is that we're now involved in the data center. There's a lot of workloads that are being run around AI. They're all running through Arm. That's great, and we're spending a lot of investment about how to make our CPUs more energy efficient and how to solve that problem in a better way. On the flip side, basically the client-cloud model, which has existed for decades. AI is not a client model today. It's not even close. And history has taught us that ultimately everything on some level of compute, moves from the Cloud to something that's smaller. Doesn't mean that the cloud is replaced, absolutely not, but it means that some level of hybrid computing will take place. So we are spending a lot of time on that, and the reason for that is that you were mentioning earlier about Arm's footprint, 30 billion devices a year, 300 billion devices, etc. etc. But what that means is pretty much every compute device that we have today, whether it's your security camera, your earbuds, your mobile phone, your PC, Arm is inside. We believe that all those devices are going to run AI, obviously, but to what extent will they run the major workload of AI, how much of that can be done locally versus offsetting the cloud? How much of that inference takes place locally versus having to go back to the cloud every time? So we are spending a lot of time on that. By the way, it's a great time to be in technology because it's a very hard problem to solve, right? AI is heavily compute intensive. It's heavily memory bandwidth intensive. It takes a lot of energy, sucks up a lot of power. Those are all things you hate when you're trying to build small devices that have to run off batteries and, oh, by the way, these devices also have to run the operating system that they did in the past. On the flip side, I think there's incredible opportunity for another huge shift in our personal devices. And if you think about your mobile phone today, your mobile phone is a ubiquitous device. The generation can't live without it, but it was only invented in 2008 so it's not that old of a device. It's basically a pull device. You have to go into it to pull information everywhere you can. Your phone is full and littered with apps. Imagine a world that a lot of the things that you need get pushed to you. Now I'm not saying the phone goes away and maybe the form factor doesn't, but there could be other things that supplant that, that AI pushes to. You're on holiday and you're trying to figure out a map and where your dinner reservations are. It just knows, in other words, it knows you're lost, right? It knows you don't know, actually, where the restaurant is. It actually knows you don't know where the tour bus is. And that information whether it's through a wearable, whether it's through something in your ears, gets pushed to you, versus having your head down and having to look at the phone. So I'm hoping that there's a generation of not heads down anymore, where it's a bit of a heads up. And I think that's actually quite possible.

FJ (15:58)
And honestly, I don't think we're too far from that model that you're describing.

RH (16:04)
I don't think we're too far, and my intent is that Arm is the center of it, because we're running those compute workloads already locally. So what we're really looking to put effort and energy into is how to make that happen on the Arm platform.

FJ (16:18)
So this notion of edge computing becomes much more of a reality now, combined with AI, that makes sense. You brought up energy, and I want to just touch on it. You briefly alluded to what Arm is doing in this context. You actually, you've been quoted saying that companies need to rethink everything to tackle energy efficiency. We all know that energy efficiency, sustainability, are topics that you know, the entire industry is concerned about, obviously, not only in terms of energy production, but also the impact potentially on the environment and so on. We have to be cognizant of it. Beyond the specific things that Arm is doing, what are your thoughts about energy efficiency? I mean, these data centers are going to consume more and more energy. We know this. And some data shows that in the United States, from 2% of our electricity quickly, within a few years, about 12% of electricity is going to be going toward these data centers that are powering essentially AI systems. What are your thoughts on it? What words of wisdom do you have about the issue of energy and AI for us?

RH (17:19)
Yeah, it's a great question, and, on one hand, the industry got a respite, maybe not in a good way, from the standpoint of sustainability in that in the previous administration, getting access or getting permitting for new data centers was pretty hard, and a lot of companies had edicts where they were going to be carbon neutral by sometime near the end of the decade, which is really hard to do, by the way. But I think in the long term for the planet, it has to be solved. Ultimately, the rate of growth of these data centers and what's required, it's not sustainable, which again, brings me back to the role I want Arm to play is moving these things to edge devices, where you're not having to run all these giant workloads in the cloud, where you can run things in a more efficient way at the edge.

FJ (18:09)
You know, I think there's no doubt that not only we need to invest, not just in data centers, but we need to invest in the energy infrastructure of the country, the world actually, to be able to sustain it. And then at the same time, lots of research that needs to be done, to look at the potential environmental, climate impact of, and also double down on research that allows us to move down the path of decarbonization, move down the path of much, much more efficient systems that we have, not just computational resources from that point of view, but also the physical systems that we have that rely on digital technologies to be much more efficient.

RH (18:50)
And I think that's a huge opportunity here, at universities, because it’s not the kind of thing corporate is very good at. The corporate world is working on roadmaps. We're working with supply chains. We're working with lots of different people inside the ecosystem. And if you're a young startup company working in these areas, it's hard, because how do you think about the productization opportunities around that? So I think the research done by universities in this space around sustainability and energy efficiency, it's a huge, huge opportunity for it.

FJ (19:21)
I am so glad you mentioned that, because here at Carnegie Mellon.

RH (19:25)
That wasn't a setup. That wasn’t a setup.

FJ (19:26)
You knew that was coming. But candidly, I think absolutely, we need to have a very expansive research agenda for energy. I do want to put a plug in about three dozen of my faculty colleagues, right before the summit that we had, got together and in fact, developed a couple of dozen white papers on issues surrounding energy innovation and AI, covering a range of topics, from, for example, how do you cut energy demand? How do you use, essentially, AI to develop more sustainable, more energy efficient systems? To the issue that I mentioned about decarbonization of various industries, from construction to steel to transportation, and also issues surrounding accelerating development of innovative new solutions, including new materials that can be used for batteries and so on. So the range of topics that you can touch, as far as energy is concerned, and AI innovation is almost limitless. I want to go back to innovation and Arm. You know, when technology is moving so fast, when the speed of technology, and we have to admit that, in fact, again, people who are of a generation during the early 2000s, we thought the digital revolution - we thought the internet technology and so on, is moving so fast, how is society going to keep up with it? Well, it turns out today, in fact, we've seen a much higher acceleration of emerging technologies, whether it's quantum, whether it's AI, whether it's wireless technologies. How is Arm prioritizing and managing the trade-off between the speed of innovation, and being there to serve your customers and so on, engineering quality and at the same time, profitability of the business itself. How do you tackle this?

RH (22:21)
Yeah, that's a topic we talk about a lot inside the company, and we were chatting earlier about my NVIDIA time. I think one of the things that I learned probably most from working with Jensen was speed. And quality, obviously, you don't want to compromise on and ultimately, as CEO of a public company, you are accountable to the shareholders in terms of financials and profitability. But speed makes up for a lot of sins. What do I mean by that? The pace of innovation is so fast that ultimately, mistakes are going to be made relative to the risks you take and how you ultimately decide what to invest in and what not to invest in. To be able to move very, very quickly and decide what your priorities are, but more importantly, when that priority is no longer a priority, is really important. Case in point, we do long range planning. We have three year plans, five-year plans, 10-year plans. We rip them up all the time. All the time. There's no sacred cow relative to, we'll do an annual planning cycle, and two months later, it might be completely changed, because the world has changed. And I am a very big believer of a mix of long range planning, but being able to pivot in the moment. So I've been the CEO for three and a half years, and that was one of the very first things that I talked about when I became CEO, and I had a slide that said, “if things feel comfortable to you, you're not going too fast.” I don't want people to have a sense of comfort in terms of where things are, because once you kind of have a sense of comfort that you're going to speed limit, people will pass you, And 40 years of technology innovation, and you can look at the great companies that sat on top of the world, whether it's Nokia or Blackberry and some of these companies, and then people say, ‘Well, what's wrong?’ Kind of got too late to pivot. So back at Arm, that's the highest priority I put. I drive my people a little crazy with it, but that's the priority I put.

FJ (23:21)
You mentioned Jensen a couple of times. In an interview that you did, you referred to Jensen as a mentor, a friend and a former boss. So when you think about your relationship with Jensen, what are some of the lessons that you learned from him during your career, and how has that influenced your leadership style at Arm in particular.

RH 23:45
So for those who have probably watched Jensen's GTC keynotes, they are very long and they are very in depth. He has an incredible command of the technical depth, and knows his subject matter really deeply. And that's what I learned from him in terms of working with him, and that I instill. So here's a story. I had been there maybe three or four months, and he wanted to do a business review of my business. And classic big companies, you want to do a business review. You put together a bunch of slides. Your business analyst puts together a bunch of financials. You have some people helping you with the technical specs and the road map. So you’ve just got a mountain of data that you know, no matter what the CEO asks you, you know he's not gonna stump you and you usually come into the room with three or four of your team members. So he sends the invite, and it's just me, and I'm thinking, “well, wait a minute, why am I the only person in the room?” And I remember checking with his EA and said, “no, Jensen just wants you.” And I remember this so vividly Farnam, when I got up to his office, he said, “why'd you bring your computer?” I just didn't know what to say. He said, “There's a whiteboard there, go describe your business.”

FJ (24:54)
Wow. And so what did you do?

RH (24:59)
That's a very classic question. Where would you even start? And by the way, that is the mastery of that question, right? “Go describe your business.” And what he wanted to try to understand in first principle is how you think, to teach you to think about what's important. And it's very liberating when you're on a whiteboard with nobody on your team around you, to try to explain to the CEO of the company. And I was running – at that time, I was running a big business for him, and he would say, “you are the CEO of this business. You need to tell me what's important.” And by the way, he did that with everybody. But back to the example of the GTC, and now how I operate, it was “know all the details, be able to talk about your strategy extemporaneously.” And the key thing, this was probably the biggest learning, and he used to say this all the time, and my team, I think, can attest to this: “Your strategy is not what you say. Your strategy is what you do, your actions, how you're actually leading your team, how you're communicating. That is your strategy.” You may say what it is, and you may put a bunch of bullet points together, but if you have to think about conjuring that up, it hasn't come to you naturally. So that was the biggest learning. I do the same thing with my team now too, by the way.

FJ (26:11)
I'm going to try this on my leadership team next week. Folks, you know, I know you agree with me that I can sit here and listen to Rene talk about not only his personal and professional journey, but also his perspective about the industry, about Arm and also his leadership philosophy. But I did promise him that we will turn the table, and he can ask me a few questions, and then I have one final question for Rene before we wrap up.

RH (26:37)
One of the things that, and it comes up a lot in terms of when we think about AI, you know, inside the company. And I'll just give an example, that we're starting to use it a lot inside Arm, and we're using it in the GNA functions, finance, we use it in legal, but we're starting to use it inside engineering. And again, Arm's products are basically the building blocks for doing chip design. We don't build the chip, but we basically design the intellectual property. So we create the RTL, we do the verification, we do all the documentation. So we're a virtual chip company, if you will. There's a lot of work involved with that, and the company's about 80-85% engineers. We're now starting to use these tools a lot for work that our graduate, what we call in the UK, graduate students, would have done, stuff that's like year zero through four, and are even to the point where our head of people came to me and said, “Hey, I think we have to probably have a conversation somewhere down the road of, do we hire as many graduate engineers as we have and/or what do we have them do if a lot of the work that they were being trained on is being done by AI?” Which then leads me to, how do universities help us in this? And what are the roles of universities, and what happens with AI in terms of developing our next great engineers?

FJ (28:04)
That's really a great question, and I know that this is something not only CMU is intensely focused on, but I know many academic institutions, particularly research universities, are looking at this issue. If you look at some of the studies that are out there, six in 10 business leaders are saying AI is going to transform their organization within a few years. There’s data that shows that 70% of the skills that are required for average jobs will change within the next five to 10 years, and a shift that's in fact, fueled by AI. So when you step back and think about it, the question is, when a student comes to Carnegie Mellon today, what do we teach them such that they're going to be in their professional lives for another 40, 50 years, maybe even longer, depending how medicine progresses and so on. So the issue becomes, what do we do in the short term and what are we going to do in the long term? In the short term, there's no question that we need to be very intentional about bringing AI, not only teaching AI to computer scientists, as we do now. And by the way, Carnegie Mellon was the first university to have an undergraduate degree in AI in 2018, way ahead of everybody else. But also bringing AI actually to other disciplines, not in terms of teaching how to teach other disciplines, but also integrating AI into other disciplines. For example, our engineering college has developed a bunch of masters programs where AI is brought into civil engineering, chemical engineering, and different disciplines – how AI is, in fact, changing those disciplines. So we're bringing that to bear. And I think in the short term that's extremely important.

RH (29:56)
There's a lot of debate in our world whether, five years from now, 10 years from now, some of the white-collar jobs, will we need as many of them, right? Let's take engineers and scientists, putting the professional debate aside. Do you think the enrollment for those curricula at CMU looks the same five years from now? 10 years from now, do you have the same number? Do you have more? Do you have less?

FJ (30:19)
The issue that you're raising is an important one. Would engineers in 10 years, 15 years, have the same set of skills as the engineers do today? I think fundamentally that's going to change. There's no question in my mind that, fundamentally, even the scientific discovery process itself is going to change as a result of advances in technology. So what that really means is we're not just teaching students skills that they need for the next three years, five years, 10 years, but we have to teach them foundations that they're going to be used to learn new things. If you agree with my premise that the half-life of any skill is five years, that means what we really have to teach students is how to learn new things, and constantly to be able to learn new things and embrace, essentially, new technologies, new science, new disciplines and so on. The second point related to that is, we are teaching our students also many of the soft skills. I think what we have embraced at Carnegie Mellon is, while it's important to focus on disciplinary expertise, teaching students what's known as soft skills, which I think really doesn't do justice to it, problem solving, creative thinking, entrepreneurship, communications. These are the skills, when you talk to employers, they will all tell you those are as important as any skills that we teach our students. So that has become very much integrated into our curriculum. It has been very much integrated into our learning experience, not just for scientists and engineers, for students who are studying arts, for students studying social science and humanities, policy and so on. And the final thing that I want to say about all of this is that while students are going through this process, we can't forget that the traditional disciplinary silos you were asking me about would, for example, a civil engineer in 10 years or 15 years need to have the same set of knowledge as civil engineers that has today? I think likely there is some set of knowledge that's foundational that a civil engineer needs to know, or a chemical engineer needs to know, or a chemist needs to know. But what's more important is that many of the problems that we're facing and tackling are interdisciplinary in nature. In other words, not only do you need to have some disciplinary expertise, you need to be able to connect to other disciplines. So, for me, in terms of the future of education is – how do we educate students such that they're going to be very comfortable, not only being comfortable with their disciplinary expertise, but be able to connect with other disciplines, because that's going to be as important as anything else.

RH (32:52)
Makes sense. Where I think we are at Arm, I'd be very curious on your viewpoint on this. Back to the graduate example, whereas I don't think we are slowing down hiring engineers. In fact, when I view the way that we could use AI inside our company, it's to develop products faster. It's to take a product today that could take three years to develop. We could, you know, cut that in half to 18 months. The step up would be ideas, and so another way to put this is, I think as long as we have enough ideas, we're going to have a lot of jobs for people. There's no doubt about that. But to what level, how far away do you think we are with today's AI that it can truly think and invent and create?

FJ (33:37)
That is really a question of, what do we mean by creativity? What do we mean by invention? Because there are really different definitions of creativity and invention. Do I expect that our computer systems, our AI-based systems, are going to mimic human intelligence? I don't think so. See, we don't actually understand what human intelligence is. We actually don't understand what creativity is. So to suggest that somehow we can mimic it, suggests that somehow we have a model of what intelligence is, what creativity is, and we actually don't have a model for it. So I've always believed that these systems, and we've seen this over the last several decades, they will help human beings by augmenting our cognitive and physical capabilities. So we've seen quite a progression, if you notice what happened in automation and manufacturing and so on. We've seen this in physical labor, where technology and automation and AI and robotics has come to bear, such that we're augmenting human physical capabilities, such that a lot of the work that used to be laborious, to be grunt work, is going away, and it's becoming essentially handled by machines. I think the same thing is going to happen – is happening I should say, when it comes to cognitive issues, when it comes to creativity, as we call it creativity. So a lot of stuff that may have been sort of routine is quickly going away. When it comes to doing research, there is no reason to spend so much time to do research if you have an AI-based system that can help you accelerate that research work that you're doing so that you can then pay attention and divide your attention and put it in areas that are much, much more interesting, much, much more fulfilling, and so on. So I'm not one of those people that believes that, oh, jobs are just going away. We do know that some jobs are going to go away. A lot of jobs are going to be transformed, as I mentioned earlier, that we need to be ready for that AI-based economy. But without any doubt, a lot of new kind of skills and jobs are going to be out there that we haven't even dreamt of. So that, I think, is going to be the future of work, if you want to fast forward anywhere between five to 10 years.

RH (36:02)
Do you think the question of cognitive intelligence and how we come up with ideas is too complicated a problem for AI to figure out, or why can't AI figure that out?

FJ (36:12)
No, I don't think it's actually, this is what I was getting at, is that, I would be hesitant to say that AI-based systems can mimic human intelligence. By the way, there are people who disagree with this. What I was really getting at is that, if we can't model human intelligence, and we can't quite model human creativity, it'll be hard to say that we have essentially replicated it. However, the question that you're asking is a fascinating one: can these AI based systems help us come up with new ideas, new concepts, that potentially human beings could come up with, or perhaps human beings cannot come up with? The answer to that is absolutely yes. There's no question in my mind that the entire scientific discovery itself, process, it's being turned upside down as a result of access to AI and as a result of access to technology, and it's going to transform the way we do discovery, it’s going to transform the way we do research. It's going to transform, in fact, even in a way that we take innovations or inventions, from research and prototype, to translation, to actual commercial systems. I think all of that is going to be impacted by AI-based systems, because they're going to help human beings actually do a far better job than we've done. Can we come up with new ideas based on just AI based systems? We're already doing it. I think it's already happening, and we're seeing it, and I believe that is going to be an acceleration of it moving forward.

RH (37:45)
Do you see an opportunity, as President of Carnegie Mellon to harness that in any way, to accelerate how you do research and teaching here?

FJ (37:53)
Absolutely, and we're actually doing that today at CMU. Here's the way we were approaching actually scientific discovery. We've actually created a lab. It's a wet lab, essentially, an experimental environment, where we're automating, essentially wet lab experiments remotely. So a chemist, a biologist, a material scientist, they can do essentially, should be able to do their experiment remotely by having access to hundreds of essentially potential scientific equipment. Now imagine taking that, Rene, being able to do experiments remotely and augmenting that with machine learning and AI based system such that you can essentially reduce the search space that you're searching for a potential new drug, for a potential essentially new, essentially, discovery, if you're able essentially to bring AI and machine learning and add that with automation and bring all of that together, which is really bringing simulation-based systems, if you will, with physical essentially, manifestation of this, and bringing all that together. I believe we can accelerate scientific discovery. I think we're going to come up with new things that are going to be totally revolutionary that we have not thought of. Of course, this will have impact in life sciences, in medicine, obviously, in other areas, from new materials to new battery systems and energies and so on.

RH (39:18)
Amazing, amazing. I think, Farnam, since this is your lecture series, you get the last question.

FJ (39:24)
Alright, thank you. What I want to pose to you as our last question is, if you go back a few decades and you put yourself in a position of students, what are one or two pieces of advice do you have?

RH (39:37)
I would say, experiment often and be completely comfortable with failing. Mistakes are made, we're human, and Farnam, I'm sure, can speak as well – when you get longer in your career, the mistakes you make are the lessons that you learn, and they move you forward. So I would say, experiment often and be happy with failure, because that will give you the advancement that you're going to really deserve and want.

FJ (40:01)
Well with that, please join me in thanking our distinguished speaker today, Rene Haas, CEO of Arm. Rene, thank you for being here.

RH (40:09)
Thank you. Thank you.

RH (40:19)
Thanks for listening to this month’s episode of Tech Unheard. We’ll be back in the studio next month for another look behind the boardroom door. To be sure you don’t miss new episodes, follow Tech Unheard wherever you get your podcasts. Until then, Tech Unheard is a custom podcast series from Arm and National Public Media. And I’m Arm CEO, Rene Haas. Thanks for listening to Tech Unheard.

Credits (40:44)
Arm Tech Unheard is a custom podcast series from Arm and National Public Media executive producers Erica Osher and Shannon Boerner, project manager Colin Harden, creative lead producer Isabel Robertson, editors Andrew Merriweather and Kelly Drake composer, Aaron Levison. Arm production contributors include Ami Badani, Claudia Brandon, Simon Jared, Jonathan Armstrong, Ben Webdell, Sofia McKenzie, Kristen Ray, and Saumil Shah. Tech Unheard is hosted by Arm Chief Executive Officer Rene Haas.

Rene Haas (00:07):
Welcome to Tech Unheard, the podcast that takes you behind the scenes of the most exciting developments in technology. I'm Rene Haas, your host, and CEO of Arm. Today I'm joined by the legendary musician Peter Gabriel. Listeners will know him for hits like Solsbury Hill and Sledgehammer, as well as his commitment to activism and innovation. Peter has long been an early adopter of new technologies in his music, having been on the front edge of digital recording, download, and distribution tech. He's also used innovative technology and other spheres, including the use of cameras and the internet to mitigate human rights abuses with a charity witness, always pushing the boundaries of technology and creative work. Peter ran a competition with Stability AI a couple of years ago, asking artists to submit AI animated music videos inspired by and set to his music. Peter Gabriel, welcome to Tech Unheard.

Peter Gabriel (00:59):
Thank you very much.

Rene Haas (01:00):
It is a pleasure to have you here. We're actually in the English countryside at the Founders' Forum, which is my first time here, but we were chatting earlier. This is your 15th or 16th time to this event.

Peter Gabriel (01:12):
I think I was invited to the second one, and I have no idea why, but I had a good time and I've been enjoying coming ever since.

Rene Haas (01:20):
For you personally, what do you get out of coming to something like this?

Peter Gabriel (01:24):
Well, my dad was an electrical engineer inventor, so I think that's the starting point from my interest, but it just feels like there's a lot of smart people who are going to influence what happens in the world, and a lot of the stuff they're generating is optimistic and positive. So I come away usually excited by some of the stuff I've learned. So for a curious mind, it's a wonderful place.

Rene Haas (01:51):
Yeah. We'll touch on a few things that we saw because there are actually some very, very fascinating demos here. But you mentioned your dad was an electrical engineer, and that's my background, and my dad was a scientist more into physics and chemistry, an inventor. But I'm curious, your dad being an electrical engineer, what were the kind of problems he worked on?

Peter Gabriel (02:11):
Well, I think in the war, he was part of a team that before radar, Germans used to have a beam the planes would travel along, and so they found a way to bend the beam so that the planes ended up dropping the bombs into the ocean rather than on land.

Rene Haas (02:30):
Oh my goodness.

Peter Gabriel (02:30):
So that was a useful thing. And then he got involved with cables, and I think he ran the, well, I know he ran the first experiment running tv, fiber optic cable.

Rene Haas (02:43):
Oh my goodness.

Peter Gabriel (02:45):
Which is actually credited to an American, but his was two years prior to that.

Rene Haas (02:49):
You mentioned the beam sort of diversion for German airplanes. Was he working for a defense contractor for the UK government?

Peter Gabriel (02:58):
I think it was for the government.

Rene Haas (02:59):
Oh, for the government directly.

Peter Gabriel (03:00):
But his main company was a company called Rediffusion who mainly designed televisions, and then they created a television studio, but he also was involved in aircraft simulators that Disney bought eventually, I think for theme parks.

Rene Haas (03:17):
Yeah. So that was end-quote, real electrical engineering in terms of developing.

Peter Gabriel (03:23):
Yeah. And he had his workshop where he used to run away from the craziness of the family and make things and fix things. I inherited his skill to sort of disassemble almost anything, fixing and reassembling, I didn't get.

Rene Haas (03:37):
Well, the road less traveled/forward. My father was a scientist as I said, I had an interest in getting into media and broadcasting and TV and also in engineering, and my dad said, well, if you want to pursue the latter, we'll fund it. If you want to pursue the former, you're on your own, which made the economic decision very simple for me.

Peter Gabriel (03:57):
Funny how that works.

Rene Haas (03:58):
You didn't follow up your dad relative to the engineering background. You got into music. Why into the creative area of music versus the engineering side?

Peter Gabriel (04:05):
I never had his skills. I think I may have been a bit ADHD. I was dismal at exams, so I didn't have the smart enough exam results to get into university, and I was passionate about music, and it felt like all the pent-up emotion and frustration that I had, I could pour into the music and it felt I started off as a drummer so I could hit things.

Rene Haas (04:31):
Well, it's interesting because - well, a few things. First off, probably if you grew up in this era, being not very good at exams would not have actually been an inhibitor to getting into this field, because clearly there's been a lot of people who have been very successful who weren't great academic and test takers. But on the second hand, and I've noticed this in my engineering career, there are many, many engineers that I've worked with who are musicians and very good ones. And there's something about how music is created that is very analogous in some ways to the way engineering problems are solved. Do you have a view on that?

Peter Gabriel (05:04):
I do. I think it's probably something to do with the way the brain is structured because it also, for maths and medicine - I mean, the first band I was a drummer in was not a very good jazz band, but it was all doctors. But I think somehow maybe it's patterns and organization, that some sort of skillset, and - I mean, I hate this division. You know, I think arts and science should be open to everybody, and we should use AI in part to allow us all to become artists and scientists and self-generate these creative worlds around us. So I hate people limiting themselves. I think fear determines many, many human decisions and shouldn't, but we need to pump ourselves up in various ways in order to break through the fear.

Rene Haas (06:01):
And there's a lot. I want to ask you about that in terms of AI and the brain and how we make decisions. But when you got into music, in terms of your creativity, was it something that you would write down what it is exactly you were going to try to compose? Or did the ideas come to you somewhat in an analog or random fashion relative to how you constructed your ultimate output? I'm curious you know, engineers start with a bit of the ‘what problem am I solving’ and work backwards relative to the logic required. In music, when you're creating something for you personally, how does that come about?

Peter Gabriel (06:36):
Well, I have a theory that there are two types of creative energy, Energy A, which is more analytical, which may be mostly computer-assisted, and then Energy Z, which is more intuitive and zen energy. And so for me, an ultimate creative structure would be different layers. So perhaps you map a rhythm in layer one and you improvise something, and then you use whatever brain power can assist you to improve it and focus on details. And then the next level might be timbre, next level melody. That's the theory. I mean, often it just happens in a messy random way. And I remember an interesting conversation with George Martin, because he would say, what I do, I have a clear picture and I know how to get there. For me, it's much messier. You know, I throw a lot of shit against the world until something sticks, and I love introducing the random, and I think that's a key part of design is to get a randomizer in there that's really opening things up.

Rene Haas (07:52):
Yeah. And I think when I think about AI and I think about AI relative to our industry, it's much more the George Martin aspect. In other words, it's logical, it's a work back. If you look at the way large language models work today, they work off of a known set of answers and then test and test and test and test until the right input comes back. The zen part that you just created, the throwing shit against the wall, then I wonder whether AI can ever accurately capture, and that gets into really the bioscience of the neurons inside the brain that my brain is 20 watts, your brain is 20 watts. The way your neurons fire to create something is completely unique to you. I could never do it. And I wonder whether you think AI could ever achieve that?

Peter Gabriel (08:41):
I'm certain it will. Yeah. No, I really think that, you know when-

Rene Haas (08:45):
Why are you so certain?

Peter Gabriel (08:46):
Well, I mean, maybe it's just smarter algorithms that are better generating, but I mean, there's a lot of random elements in the world, which could be inspiration if you like, for the algorithms. You know, I can't see any job that in the future can't be better done either with the help of AI- obviously person-to-person skin-to-skin, nursing or whatever else, these are harder things to crack.

Rene Haas (09:15):
So you believe that AI will ultimately get pretty good at creativity and invention?

Peter Gabriel (09:19):
Yeah, I think it's absurd that it wouldn't.

Rene Haas (09:23):
Yeah, there's a lot of healthy debate on that topic, just relative to can AI solve problems? The answer is not known. And I agree with you, I think, and we're already seeing this, right? There's the aspects of artificial intelligence that are either augmenting jobs or accelerating jobs, and with every technological innovation replace some jobs. But the creativity and invention and innovation, which is uniquely human and uniquely patterned to how our brains operate, that to me will be quite a fascinating bridge to cross when we cross it.

Peter Gabriel (09:56):
Yeah, no, it may take a little while, but I know Bill Joy a little bit, and he's working with his son on a systems-based approach to AI using fractals and patterns and trying to, I guess reverse-engineer, I'll get it wrong if I try and describe it, but it just senses that maybe rather than going from leaf to leaf, it might understand the branches and the trunk and the roots a bit. And if that materializes, maybe that would help in this process. But it then brings up all sorts of societal questions. We tend to go towards the most effective and cheapest solutions to all problems. So where does humanity fit in that? And the universal basic income is one possible solution, but we don't have the resources to jump to that. So I think this is a very interesting and potentially awkward transition that we're making.

Rene Haas (10:54):
We're in the midst of some things that - and you and I are roughly the same age - but I did not think that we would see in our lifetime. This is one of these things, I always thought that a few generations would have to worry about this problem, but not ourselves. But now it feels like we're front and center of it, which really raises a lot of fascinating questions, as you said, in terms of moral, social, societal, in terms of where things are going to go with this.

Peter Gabriel (11:20):
One project we were worked hard on was to create a thing called theelders.org. Richard Branson and I went to Mandela and tried to get former world leaders working together so that they would have a currency that wasn't based on economic, political, or military power, but just on moral authority.

Rene Haas (11:43):
Oh my gosh.

Peter Gabriel (11:44):
So that's just one thing, but there's many people working on different projects around the world, and I'm passionate that we need to find a way right now of connecting all the people that have optimism and hope, because, you know, I still believe you can go anywhere in the world and you'll find kindness, compassion, generosity. But we don't have a means of harvesting those. Whereas right now, hatred and division, we can harvest and we can make particular people very powerful through that process. So there needs to be a counter-movement and I think there's a growing group of people looking at ways whether you can create global passports, global citizenry, global commons, all sorts of ideas that I think could bring people in. And then whatever amount of military that gets thrown at this, ideas can't be killed off while there's one person left standing,

Rene Haas (12:42):
I was not familiar with theelders.org. Is it still - you said Mandela.

Peter Gabriel (12:46):
Yeah, still going. And again, I think we are looking and hoping that there'll be ways that, I mean, they have successfully influenced quite a few difficult situations, but it was an example.

Rene Haas(13:00):
Interesting. What are some? I'm just kind of curious.

Peter Gabriel (13:02):

Well, I think where there's been war about to break out, I mean a couple of places in Africa and so on, I think they've gone in. Tutu was the chair for a while and Kofi Annan afterwards, and then they've been able to go to Myanmar and Cyprus, places that have tensions still. I mean, Israel Palestine is still - I don’t know how effective they were there, but they certainly let their voices get heard. And the idea was, you know, Mandela was quite clear at the opening, he said, if you can't see and feel this in the village, you're not doing the right thing. They had one program against child marriage, which has changed laws in a number of countries. So there's small incremental things. But if there was a movement which could unite various people working on climate or indigenous rights, land rights, all sorts of things that could get connected.

Rene Haas (14:00):
Are governments involved in it, or?

Peter Gabriel (14:02):
No, and anyone who's still active in politics is not allowed to be an actor. But the Elders select themselves.

Rene Haas (14:09):
Gotcha. I get there's not a bad thing to kind of have a self-selecting body in terms of that sense.

Peter Gabriel (14:15):
Yeah, but I mean, but a dream might've been that may still happen eventually, the people of the world try to elect people who they think have made remarkable difference with their lives. And obviously we've had a fair number of Nobel Peace Prize winners in amongst the Elders. So but it's an example of an initiative, which, you know, I'd love to see connected with people at the bottom, so that it's-

Rene Haas (14:41):
That's fantastic.

Peter Gabriel (14:42):
Well, yeah, but I think there's a lot of people doing things, but we just need to sew it all together.

Rene Haas (14:49):
You've been involved in so many interesting, fascinating things. We could spend so much time. I want to go back to the founder’s forum we're at, you and I just walked through the demo tent and we saw something interesting in terms of a regenerative electronic process. We drank some algae together. Good for the gut. I'm not sure, I'm not sure, might be last algae drink for quite a while. You said your dad was actually involved in something kind of interesting, similar in terms of -

Peter Gabriel (15:15):
Well, he wasn't involved at all. He was a consumer, and I think mainly he bought sort of plasma-creating electronic device that would, I think he was hoping cure baldness. But it was, I mean, I've looked at, I think it's from the 1950s or ‘40s, but it's got a whole list of therapies, which I'm not sure how many have the evidence behind them, but I'm passionate about energy therapy in the sense that the way pharma is structured at the moment, we are going to serve the wealthy once again, but we're not - maybe certain exceptions - going to reach billions of people around the world. However, and this is where your world comes in, I think if you can get consumer electronics generating infrared healing, ultrasound, better understanding of what's happening with the electricity in the body and magnetic fields response that we can maybe get directly to individual cells and get something that you then connect to your phone and could provide high-tech healthcare to billions of people at an affordable price.

Rene Haas (16:21):
Yeah, you and I were chatting a little bit about this, and I've become familiar recently with energy therapy, and I'm a little bit of an old school guy, just from the standpoint of if it's not covered on their insurance and covered in your med plan, is it a legitimate type of procedure? But people who I know have used it-

Peter Gabriel (16:40):

Rene Haas (16:41):
Some very close to me, and they swear by it. And it makes a gigantic difference in their lives.

Peter Gabriel (16:48):
And there's now the evidence there. There is the good science.

Rene Haas (16:51):
So what's the blocker behind this?

Peter Gabriel (16:54):
Well, I think you need a cultural shift towards healthcare that the old days where you got sick and who do we go to? We go to our parents and say, an expert. We hand over responsibility to someone else, fix us, please. But it doesn't work like that, particularly now, we've got to be collaborators. Both sides asking questions. Not one side who knows and the other who receives.

Rene Haas (17:19):
It's an area where Arm technology is such a natural fit. Arm is in a lot of medical devices today, simple devices, home blood pressure monitors and things of that nature. All of that work that's being done is all happening on an Arm-based solution. And we've had a lot of people inside the company for years have been passionate about addressing this field, we've run, we tend to run into a lot of regulatory issues. Just simply for the people who are the device manufacturers commercially, they just look at the amount of time it takes to develop such device and the cost, and they run out of the gumption to continue with it. But it's low hanging fruit, quite frankly.

Peter Gabriel (17:56):
And it's less regulated than drugs. So that there is a way maybe through consumer electronics of getting some stuff tested that can just be made available. And then the science and clinical trials can take place afterwards as long as people aren't dying in the meantime.

Rene Haas(18:14):
Yeah, and I think back to the drug industry, they've got their own motivations in terms of how pharma is set up. It's a highly regulated industry. It's a very expensive industry. And alternatives are absolutely needed.

Peter Gabriel (18:26):
And I just think there's so many opportunities and you're beginning to see startups in this area, and there is growing evidence.

Rene Haas (18:33):
Yeah. No, we just ran into a couple interesting ones just in the demo area. Switching gears for a minute, I'm going to talk about Stability AI. Yeah and I read about your interesting work in terms of creativity there around and music and such, or maybe video. Can you tell us a little bit about what that is and how you got involved with it?

Peter Gabriel (18:49):
Well, I just think I love any creative tools that suddenly allow ordinary people to do exceptional things. And so AI has definitely given that opportunity both in video imagery and music. Now, I was very happy to just open up my music for people to experiment on. I mean, there are still some ethical issues about training models and all the rest and who it takes from, but - you know -

Rene Haas (19:17):
What does it do exactly?

Peter Gabriel (19:19):
Well, no, this was just using prompts to generate, I mean, still it's mainly a technical nerd community doing these things. But we also set up a thing, because at the moment, if a video goes onto YouTube, the musician or the songwriter gets most of the income outside of the tech company, and we felt it should be - if it's a good visual thing. But so we have a site we call 50 50, which we're just experimenting with, which will divide the income, which we feel should be larger to the creative side.

Rene Haas (19:59):
A bit of a democratization then, if you will.

Peter Gabriel (20:00):
Yeah, we're trying a little bit, but I just think enthusiasm is what I've always been drawn to. And if people want to try things, and I think music is going that way too with its sort of evolutionary approach to music. So we had a slogan, it's the process, not the product. And that if you can invite people in on the process so that they become co-creators with you and some will bother, and I don't want to bother on a lot of occasions, but when I do, I want to have the chance of getting in there and sort of growing this garden that I can help design.

Rene Haas (20:37):
Not talking about the music industry too deeply from an economic standpoint, but if you look at where it is today in terms of how people consume music, are we in a good place? Are we in a healthy place relative to both thinking about it as an artist and then also just as how the population consumes music?

Peter Gabriel (20:56):
Well, I love the idea that anyone can get anything, and that's brilliant. But it's pushed artists' rights back about 50 years. But we had a competitor to Spotify in the early days, we were quite often early in the field.

Rene Haas (21:12):
You were very early.

Peter Gabriel (21:13):
Yeah, we had a music distribution thing two years before Apple, and we had this somewhat similar thing called WE 7, but they actually did it better. But with this one, we worked out we were paying artists 10 times as much as they were getting from Spotify for per stream. So they're very successful and we sold ours on and not so successful.

Rene Haas (21:37):
So OD two, pre iTunes, pre iTunes, post Napster.

Peter Gabriel (21:42):
I guess it was - I can't remember. It was around that time. Yeah.

Rene Haas (21:45):
Yeah. That is quite a bit ahead of its time.

Peter Gabriel (21:49):
Yeah, I think we were but - which was often my dad's problem.

Rene Haas (21:53):
Yeah, I mean, I think the statute of limitations for illegal downloads on Napster has probably passed. Yeah, but I remember when Napster kicked off, it was a little bit like Spotify now in the sense of, 'oh my gosh, everything's available and how fast can you download everything?' And then obviously iTunes put it in a different place. But the industry changed forever in terms of distribution of music.

Peter Gabriel (22:16):
Which is brilliant, and Spotify was very well designed in a way that we weren't, but I think they could afford to be more generous now they've taken dominance.

Rene Haas (22:27):
Yeah, no, I think, like you said, as a consumer of music, and I love music, it is wonderful that you can get everything at any time, anytime that you want. But it feels like all that consumption can't be good for the artists.

Peter Gabriel (22:39):
Well, this is a question for you then is: should prompts be transparent?

Rene Haas (22:46):
Yeah.

Peter Gabriel (22:46):
Yeah. I think if you've got a history that goes with any piece of software of origination and influence, then maybe micropayments. I mean, there's no real excuse why we shouldn't be able to generate micropayments.

Rene Haas (23:01):
Easily. Right. I mean, selfishly, I must've done a bunch of Spotify hits on games without Frontiers and Sledgehammer and Lamb Lies on Broadway. And I'm thinking somebody should be benefiting from this other than me just sort of slipping it. I hope Peter's getting a little piece of what's going on here.

Peter Gabriel (23:17):
Yeah, a small piece I'm sure. It's better now, but it's still, I would say it has a long way to go. Old, well-established artists, we've done very well. We were there in the financial heyday of the music business. But for young artists and minority interest artists, this is really critical because live work is very hard to get.

Rene Haas (23:38):
So one of the things that we talk about in my industry is what jobs will AI replace? We're already seeing with things like agents and chatbots and things of that nature where white collar jobs can be replaced. And on one level, you look at it and say, what's to worry about here? We've seen that with every technological evolution. There are jobs that go away, and things get more productive. Back to the discussion we had about invention and creativity, we maybe get to a world where music can be generated artificially, and we're kind of there now.

Peter Gabriel (24:12):
We're there.

Rene Haas (24:12):
But how do you feel about that, both ethically, technologically, morally, in terms of, is this a bridge too far?

Peter Gabriel (24:21):
No. You can't. You know, it's like King Canute in the waves. You can't stop it. You've just got to work with it and find your corner. So I think that's the only way to compete with AI is to work with it.

Rene Haas (24:33):
Do you have a viewpoint though? I mean, will you get to a point where there will be music created by AI with unique vocals and unique backgrounds that will say, this sounds pretty darn good. I don't really care whether it was created by a man or machine?

Peter Gabriel (24:44):
Well, as a producer always used to say, 'give me a music business without any bloody musicians', because that’s where all the problems come. And we're going that way. There's this wonderful designer called Gaetano Pesce, and he used to say that beauty in the future will lie in the imperfection, and humanity is loaded with imperfection.

Rene Haas (25:07):
I wonder. It's folks who say, we ended up having a chat with somebody. We had a very, very nice restaurant in New York a couple of days ago, and we were talking about could robots replace the chef? And my patron lunch was saying, there's just no way people will want to know their food has been created by a person. And I said, well, but we don't know who's behind the kitchen. No, the front of the house will always be people, but behind the curtain, the food tastes great. I don't know any difference. And he said, 'No, no, no. People will come to know it's made by humans.' I wonder if the same is through with music. People will say-

Peter Gabriel (25:42):
No, well I think - there is already a robot chef, and I think that it's only going to get increasingly better. And same with music. So I think the fringe is safer than the mainstream because where the mainstream is where the money is, and therefore, that's where AI will first focus.

Rene Haas (26:03):
And I think the other thing it can't replace is obviously live performances.

Peter Gabriel (26:09):
That's a lot harder. But I mean, I went to the ABBA show and you know -

Rene Haas (26:16):
Humanoids were up there?

Peter Gabriel (26:17):
Yeah, well, Abba Voyage, it's all sort of virtual characters.

Rene Haas (26:21):
Oh, really? Yeah. I've heard about that. Is that a good show?

Peter Gabriel (26:24):
They do it really well. I mean, I think there’s things for all of us to learn, but it's - as the starting point - they've done it really well. And there is already, you know, Elvis things and Michael Jackson shows coming along so,

Rene Haas (26:37):
Done - like a virtual?

Peter Gabriel (26:39):
Yeah, virtually, and that's only going to get better. So, you know I don't think anything is safe from AI. And I think that it amplifies and accelerates everything, and humans are not amplifying and accelerating their reaction and anticipation.

Rene Haas (26:57):
So now you've got me very curious. So this ABBA show, it's-

Peter Gabriel (27:02):
Yeah, you can see it in London.

Rene Haas (27:03):
Yeah. No, I've heard of it, but it's a virtual-

Peter Gabriel (27:05):
Yeah, it's all virtual. And no, I used to make a joke about it because you know I think they've taken a younger version of themselves. And so I'd said in opposition, I've made my avatar a little fatter, a little balder than the real thing.

Rene Haas (27:25):
When do I see Peter Gabriel and Phil Collins in the mid-1970s? When does that come out? I want to see that.

Peter Gabriel (27:30):
Yeah. Well, so do I. Because I think all artists want horizontal income, which is where you lie in your bed, and the money comes in.

Rene Haas (27:38):
Gotcha. I will look forward to seeing Genesis perform again in some virtual tour. Peter, it was a pleasure. No, it was a real pleasure. Thank you.

Thanks for listening to this month's episode of Tech Unheard. We'll be back next month for another look behind the boardroom door to be sure you don't miss new episodes. Follow Tech Unheard wherever you get your podcasts. Until then, Tech Unheard is a custom podcast series from Arm and National Public Media. And I'm Arm CEO Rene Haas. Thanks for listening to Tech Unheard.

Rene Haas [0:07]
Welcome to Tech Unheard, a podcast that takes you behind the scenes of the most exciting developments in technology. I'm Rene Haas, your host and CEO of Arm. Today, I'm joined by Mark Chen, Chief Research Officer at OpenAI. You might know them as the creators of ChatGPT and DALL-E. Before serving as Chief Research Officer, Mark was the Head of Frontiers Research at OpenAI, focusing on multimodal modeling and reasoning research. Mark has led the teams that created DALL-E, developed codecs, and incorporated visual perception into GPT-4. Now in his current role, Mark's goal is to push the frontier for open AI science and research.

Rene Haas [0:43]
Mark Chen, welcome to Arm. You've come all this way to meet with me. Thank you so much.

Mark Chen[0:47]
Yeah, thank you for having me.

Rene Haas [0:49]
Yeah, no, good to see you in person. You know, they say when you read autobiographies or biographies, you know, start at the beginning, because the beginning will tell you a whole heck of a lot. So maybe just starting with where you grew up, how you got into technology the way that you did.

Mark Chen [1:03]
Absolutely. So it's a really hard question to answer. You started with probably the hardest one, but I was born on the East Coast, but my parents were very nomadic, so we moved around a lot. My parents worked at Bell Labs and then they moved over to the West Coast. So I did part of my schooling in the West Coast-

Rene Haas [1:22]
In Holmdel, New Jersey?

Mark Chen [1:24]
They were in Edison. But in the Holmdel site.

Mark Chen [1:27]
Exactly, yeah. So after that, you know, my dad got the startup itch, he moved over to California. We were there for a couple of years and then we went back to Taiwan after that. So that's where I did my high school and part of middle school.

Rene Haas [1:39]
Oh, my gosh.

Mark Chen [1:39]
Yeah.

Rene Haas [1:40]
How old were you when you moved?

Mark Chen [1:41]
I think, uh, probably 12 or 13 years old.

Rene Haas [1:44]
So coming from the U.S. public school system, when you got into Taiwan, you're like, Oh, my God.

Mark Chen [1:48]
Yeah, no. It was rebellion for half a year. But I think there is a kind of love where, you know, it starts as hate and you know, after six months, I really love Taiwan. You know, it's just the center of the chip ecosystem. So yeah.

Rene Haas [2:03]
For sure. I am quite curious about coming from the- was it a public school or private school in the U.S.?

Mark Chen [2:08]
It was a public school.

Rene Haas[2:10]
So coming from the U.S. public school system at age 12 or 13 into Taiwan, did you find like, oh, my gosh, I got to catch up?

Mark Chen[2:16]
Well, it was a little bit of a culture shock, but I already liked math and science a lot. So a lot of it was just, you know, going from more of a kind of a free-spirited teenager to an environment where everyone's wearing uniforms.

Rene Haas[2:29]
Right.

Mark Chen[2:30]
You have fairly strict, uh, kind of teaching style. So that was the big adjustment. But I think it was good to get experience from both of these worlds, right? One where it was more about kind of learning for yourself and another environment where it was about discipline.

Rene Haas[2:45]
Gotcha, gotcha. And then did you do university in Taiwan?

Mark Chen[2:49]
No. Then I went back. So I did my college at MIT. There I studied math and computer science as well. But I was a late bloomer to computer science. Yeah. I only really started towards the end - programming towards the end of my college career. One of my roommates kind of goaded me to do it. And I was like, you know, hey, you know, I think I'm going to be a mathematician, but I'll try this thing out. And of course, it's addictive right here for sure.

Rene Haas[3:13]
Yeah, yeah, yeah. So, so math and science were your proclivities, but not so much into computers until you-

Mark Chen[3:20]
Yeah. Not, not into practical programming until later in life.

Rene Haas[3:23]
Yeah. Gotcha, gotcha. And then and then after that, you know, tell us kind of what you did was your first-

Mark Chen[3:28]
Yeah. So my first career was in finance. It really was a little bit of an accident as well. So I thought I was going to go into theoretical computer science and then on a whim, in my senior year of college, I took an internship at Jane Street and it showed me kind of the appeal of working in industry. It's very pragmatic, but you still have a lot of the really exciting problems that you do studying in a more self-contained kind of academic environment. So that was exciting to me. I spent a couple of years at a hedge fund and then a couple of years at a high frequency trading firm as a partner.

Rene Haas[4:06]
We talked about this once before, but I think for the listeners it'd be quite interesting because that's not exactly what OpenAI does.

Mark Chen[4:13]
No, not at all.

Rene Haas [4:15]
But there's a lot of similarities in terms of just the way of thinking from a mathematics perspective, from what goes on with high frequency trading to AI models. What were the things that helped you bridge that from the stuff you had done in the finance world, which you did, what, for five, six years or so-

Mark Chen [4:29]
About five or six years.

Rene Haas [4:30]
To OpenAI?

Mark Chen [4:32]
Yeah. So I think the biggest thing is it teaches you rigor and experimentation. So, in the financial markets, there really is no kind of fudging the benchmarks or anything, right? You have a very hard evaluation, which is how much money your models generate. And you really have to be honest, principled, rigorous in your experimentation. I think a lot of that carries over to the science that we did at OpenAI in the early days.

Rene Haas [4:57]
Yeah. Were the traditional AI - traditional is a funny thing to say in such a new industry, but - transformer based models doing the work when you were in the finance world or how would you kind of think about neural networks relative to the thinking world versus the quant slash financial space?

Mark Chen [5:15]
Yeah, so I would model finance as about two years behind the state-of-the-art in AI at any given point in time. So I remember when it was 2017, 2018, the time that I was exiting finance, people there were starting to catch onto neural networks, they were building out their first clusters. And, yeah, I was also looking into kind of neural based modeling in the finance world, as well.

Rene Haas [5:41]
Right. And just looking now back where so much is happening in terms of obviously in the AI space, we want to talk about, how has the finance world picked up on that from what you've seen?

Mark Chen [5:49]
Yeah, so they're also all in. But I will say one dividing line is you tend to get more AGI true believers on the tech side and people on the finance side today actually are still more AGI skeptical.

Rene Haas [6:04]
When you say AGI skeptical from the finance world. What do you mean?

Mark Chen [6:07]
Right, I think it's just straight up when you feel like, hey, do you believe that AI will fundamentally transform the shape of the economy in five years? They tend to have a more pessimistic view.

Rene Haas [6:16]
And why do you think that is? It's funny you say that, because I do find that also. Not finance per se, but other industries that I interact with, who have different, either methodologies for scientific research or product development. However they do that, the skepticism bar is a lot higher than I would have intellectually anticipated. But in the finance world, why do you think that is?

Mark Chen [6:40]
I can only answer for the high frequency trading world where a lot of it is extrinsic to the pure modeling. So, I really do believe that AI can improve the modeling side of things. But there's so much on latency, there's so much about your private sources of data. So, I think there's just so many extrinsic factors of alpha that maybe AI and modeling itself just really isn't a huge part of it yet.

Rene Haas [7:05]
As I said, I run into this, you know, in other industries. And some of the things I hear back is, our problem cannot be modeled. The data sets either don't exist and/or the problem is a far more complex model than AI or AGI could ever, ever assist with. You think that applies to the thinking in the high frequency trading world?

Mark Chen [7:23]
Well, I don't know. We hear this refrain over and over again. I think you have to be in the world to really believe in it. Sometimes you just have to see the tech to start feeling the AGI, as we call it. And I think at OpenAI actually going in, I would say I was somewhat of a skeptic when it came to AGI. I think much of the world was back then and really just seeing the progress of the models seeing the capabilities that really opens your eyes to it.

Rene Haas [7:50]
Yeah. So, the story as to why you joined OpenAI, is a cool one. Why don’t you maybe share that with folks.

Mark Chen [7:56]
Yeah. Well, I think over time it has a lot to do with finance and it also has a little bit to do with AI as well. But, you know, finance, it's, I think, a hard industry for someone who wants to make impact. And what I realized, having been in finance for many years, was that the set of competitors, it's the same. Everyone gets a little bit faster. But at the end of the day, you're still competing against the same people, you're still competing on the same objectives, and extrinsic world doesn't change too much for it. And I felt like, you know, that wasn't a very satisfying way to live out the rest of my life. And it felt like I needed to make some sort of change. At the same time, I saw what happened with AlphaGo, and that was equally inspiring, scary, you know, so many feelings wrapped up in one, and I just felt like I had to really get into that. So, some of the first projects I did were in reinforcement learning, just figuring out how do I train these Deep Q-Networks to play Atari games? And once you start taking on some of these projects, it's very addictive.

Rene Haas [9:02]
But OpenAI in 2018 was a very different company, obviously, than they are now. You know, it's much, much, much smaller. Charter was a little bit different, and I think that drew you a little bit, too, to kind of what-

Mark Chen [9:13]
It did. Yeah, it did. So I joined a nonprofit at the time, and in some sense you can view that as a reaction away from finance. So, I felt like I was in a very materialistic world and I wanted to do something with impact and I do feel like a lot of early OpenAI had that mindset of we believe in transformative AI, we want to bring it about to the world, we want it to benefit humanity. And that persists to the leadership today.

Rene Haas [9:41]
Absolutely. Yeah. It is very clear that Sam is a big believer in that. So, you've been there now seven, eight years. Are you surprised at how far it's come in seven years, or do you look at it and say, you know what? There's so much more to go? I mean, how do you think about the last seven years?

Mark Chen [9:56]
I think both are simultaneously true. So, I don't think anyone could have predicted this trajectory of AI. All we could do was predict kind of what we call perplexity. So this is the ability for you to model, let’s say, human language or a measure of accuracy and some sense of being able to model human language. And you can predict for a particular scale what kind of accuracy you're going to get. But what you can't predict is the emerging capability that falls out of getting that level of accuracy. So, you know, when we got to GPT-2, it wasn't obvious that, you know, this level of perplexity would mean that you get coherent paragraphs. And then with GPT-3, not clear that that level of perplexity would give you the ability to do in context learning. And then GPT-4, you know, the ability to crush all of these college level exams. So really it's really inspiring to see all of the emerging capabilities that come out of the model.

Rene Haas [10:55]
One sense of scaling and wonder how you think about this, how much of the progress or scaling of the capability, as you say, is a function of breakthroughs of the models versus we just have access to far more compute and we can just throw more dollars at it, either whether it's chips or power or whatever it is.

Mark Chen [11:15]
I think this was studied once and clearly both factors are very important. But I think the algorithmic insights and efficiency improvements have slightly outpaced the compute contributions.

Rene Haas [12:28]
And can you say a bit more about that without giving away, no secret sauce to be given away here but, anything just that would give the indication is where the algorithms have just gotten smarter.

Mark Chen [11:37]
Right. Right. So I think today, you know, there's a sweeping generalization of, we use transformers to train language models, but transformers themselves have evolved a lot over time, too. And I can talk about some earlier work. You know, we evolved the attention patterns of transformers. So I did some work early on with Ruan at OpenAI about fact-raising attention and that that's an efficiency gain. There are other subtle things too I think where you do normalization in transformers, how you set up the aspect ratios. All of those things are things that you have to figure out and be very careful about.

Rene Haas [12:10]
Is it transformers till the end of time or do you think we get where there's an S-curve, where transformers run out of runway and there's something next?

Mark Chen [12:18]
It may be possible that something takes over, but I think the longer that transformers are here and the longer that they remain dominant, the more we co-design around them. So we will build chips that are efficient there. They become the benchmarks, right. And we will build kernels for them. So it just - the bar keeps getting higher for something to overtake transformers and it's hard to see that-

Rene Haas [12:39]
Are we there already? Because I do wonder, to your point, the model has been established, hardware architectures, you know, morph to it, scientists know it. And you may have this, this bias that starts to run in to, say, if everything is transformer-based in terms of how I develop the next data sets and models - that's how it's going to get done.

Mark Chen [13:03]
Yeah, I think transformers, they're popular because they're a really good balance of simple and expressive at the same time. So they get all of the mixing primitives that you need a highly expressive mixing primitive, but they're also very simple, which allows you to scale it and engineer it in a very fairly straightforward way without too much gymnastics. So I think it's a very well-suited architecture.

Rene Haas [13:26]
Yeah. I don't know how much you interface with potential clients, but do you hear the safety thing increasing, now, given with the capability of certainly [GPT-]4.5 and [OpenAI] o3?

Mark Chen [13:35]
Yeah, I think so. And you know, today is a world where we're moving towards AI with connectors. So these are AI models that can plug in to your email, your Google Docs, your Slack. And I think that poses real risks. People today are fairly good at jailbreaking models, too, if there's a very motivated hacker. So what is the implication, right? You could imagine that someone motivated extracts all that information away from you or they're able to launch some kind of coordinated attack.

Rene Haas [14:03]
Right, so then literally at the source code level, you could just simply put things inside the model that when those queries are requested, be rejected.

Mark Chen [14:12]
Right. Yeah. So I think there's even the possibility you allude to where, if there's a bad faith model developer, they could create a model and bake in certain behaviors into weights where there's some kind of trigger for that happening.

Rene Haas [14:27]
Yeah. No, it's interesting because I remember attending the very first AI Safety Summit which was, gosh, almost two years ago and there was a lot of talk on it then. And maybe it's just the circles I'm moving in, I haven't heard as much on it, even though the models have progressed incredibly since then. So that makes a ton of sense. You've done a lot of work, I know yourself, on multimodal and that's yet another yet extremely cool frontier. Are the architectures today well suited for multimodal in terms of, it's a big data problem, right? The data sets are really gigantic, back to the transformer piece. Just talk a little bit about multimodal and where you think that goes.

Mark Chen [15:09]
Absolutely. Yeah. This is a topic near and dear to my heart because one of the first projects I did was trying to apply transformer architectures to multimodal data sets. So one of the big papers that I'm still very proud of today is called Image GPT. And that was a proof of concept that you could do image generation using the same text transformer stack, right? And all you have to do is view images as a language in this very special vocabulary of pixels. And I think at the time, you know, it was, you had different architectures like GANs or VAEs that generated images and used transformers for text. And I think the importance of proving out this approach is it paved the path for something like DALL-E, right? Where when you have image generation, you want it to be steerable image generation. So you want text to be the language by which you specify the image you want. And in that world, if you have a different architecture for text and images, it's just not satisfying, right? You would want to be able to just throw all of your data into the same model and train it to be able to output the image. So since then, we've expanded on that idea and created [OpenAI] 4o and this is a model we launched last year and it really showed the strength of a fully multimodal integrated approach where you just throw audio, images, video, text all at the same model and you can get, you know, really emotive speech. You can get images like the image generation launch that we just did and that was so fun.

Rene Haas [16:42]
Yeah, super, super cool. But it also seemed like it’s - and I guess if you're in the world of selling power or selling data centers or selling chips, a good thing because the data set just explodes in that world.

Mark Chen [16:54]
Right. Yeah. I think there's so much intelligence locked up in multimodal data and just the ability to expand your data set by orders of magnitude through that and be able to try to figure out how to unlock that is important.

Rene Haas [17:06]
One of the things that when I talk to researchers in different fields and different areas back to this problem is too hard for AI/AGI to solve - when you think about real world problems, whether it's around chip design, whether it's around pharmaceutical research, whether it's around drug discovery, link to that. Where are areas that you look at and say these are really, really tough problems, that we're not there yet, but we can get there?

Mark Chen [17:35]
Right. So I think everything above that you mentioned falls in that class. And to back up there, this is one of the big motivations of us working on reasoning. So our reasoning models today, they give us this hammer where we can take smaller amounts of data and efficiently learn on a smaller amount of data to get the same level of capability. So, many of these verticals, like you mentioned, they're not going to have as much data as the whole Internet. And you need a technique that allows you to productively spend compute, learning all that data. And that's what reasoning gives us. It gives us a tool for doing that. So we're actually very excited for the possibility of models to get very specialized in these areas. We've already seen that with things like math and computer science, right? These models are so good at solving math problems right now and so good at, you know, coding up very difficult algorithms. And I think a lot of these techniques can be adapted to things like drug discovery, like you say.

Rene Haas [18:32]
Right. Are they good at invention? On one level, they're awfully good at solving problems that are bounded problems,

Mark Chen [18:38]
Yes.

Rene Haas [18:39]
Right, where the answer is known or there is an answer that you can get to. But what about areas where there isn't an answer today. Invention being a broad term for that.

Mark Chen [18:51]
So I have the slightly controversial claim that they're better at invention than we think. And I- the reason I think that is, you know, we've entered these models into some of the hardest algorithms, competitions in the world. And, you know, I have a lot of experience in these competitions. They're often designed to be anti-pattern problems. Basically, what makes a good problem is you can't fit it to a set of known techniques, right? Otherwise, all the contestants are going to know how to solve it. So what we find is oftentimes these models surpass our expectations on these more ad hoc, like, you have to come up with something that really doesn't fit any pattern. And it does really surprise me in some of these problems. It can be very creative and in spots where you don't expect it to be. And I think you do speak to a good gap, though. There is some sense in which they are models which take tasks and then give you responses to tasks. And you would ideally love them to propose tasks and have taste in some sense of this kind of things are hard, you know.

Rene Haas [20:02]
Can they be solved? And how much of that is we need more data and/or our models in terms of how the human brain invents based upon having either not seen something or been exposed to something?

Mark Chen [20:15]
Right. Right. And I think today, models probably already do have a sense for what intuitively is, you know, aesthetic to a human, right? And I think you can leverage that to figure out hypotheses that, you know, would be interesting to humans as well. So I think it probably has this latent notion of what innovation looks like, but we should just figure out how to tap into that.

Rene Haas [20:41]
Yeah, it's always a thing I've thought about with AGI, you know, if I make it analogous to a human brain, where the AI today is incredible - given enough data and enough time, it'll figure out any problem. Yet humans aren't always exposed to everything.

Mark Chen [20:53]
Right.

Rene Haas [20:54]
Yet somehow we figure out paths and ways to learn things, having actually never been exposed to the entire data set. And to me that to some extent everyone's got their own kind of weird definition of AGI. To me that is a bit of AGI, is being able to learn and invent without actually having all the data fed into it.

Mark Chen[21:11] Absolutely. And that's the way we're going, right? It's like we're developing more and more data efficient algorithms where you're learning how to reason and draw insights from less and less data. I think one challenge, though, that I've personally reflected on for a while is how much of invention is interpolation. And I think it's actually a higher fraction than we think, you know, having seen the math world for a while, actually, a lot of very cool results are, you know, you take one person who's a geometer or some person who’s an algebraist and they link together some patterns that they've discovered in their own fields and they bring it together. And I, I think maybe there's a lot less true pure innovation than we think.

Rene Haas [21:59]
Can AGI replace or augment entrepreneurship?

Mark Chen [22:02]
Great question. So I think it accelerates entrepreneurship and- in the sense that I think if we really do coding models right, we create some sort of interface where a human doesn't have to learn how to code, to produce something that they can ship to the world, right. If you can say, hey, I want to build an app that connects writers to drivers, right, that can just get built. You know, the model thinks for a long time it makes it robust and deploys it to the world.

Rene Haas [22:33]
So many industries have been washed out by technology advancements. And there's always the doomsayers who say this is going to eliminate jobs, etc., etc., stenographers and typewriter tools, for example, that were that were replaced. Coders are a different species, obviously. Um, and there's one axiom that says the efficiency of these models will generate ability for more coders versus another axiom that says, five years from now, we won't need as many as we've got today. Where do you lie on that?

Mark Chen [23:04]
Yeah, I mean, I just think there's a lot of demand from services that isn't met. And AI will be a point of leverage for people, for professionals especially. They're going to be able to be 2X more productive or 3X more productive now and service that many more people. I think it will drive the costs down, but maybe the demand is there at a lower cost, right. So I think it could be different for different industries, but I do think there's a lot of demand out there given a cheap enough cost.

Rene Haas [23:29]
That's kind of what I think, too. I mean, we're in a bubble from the standpoint of where we live physically and the world that revolves around AI that is Silicon Valley. So much of the world yet still has a long way to go to embrace technology, let alone AI. But the speed at which this is moving is just amazing. I mean, just looking at that, just your tools, your deep research tools and reasoning tools, the answers are not they're not textbooks anymore. They’re succinct, you're actually talking to, to a human being.

Mark Chen [24:01]
Yeah, absolutely. I feel like really the bar for deep research is hours of economically useful work. And I think it lives up to that. We've gotten testimonials from people in biology, for instance, who say when it comes to drug discovery, I have a panel of scientists and having them in a room for a couple of hours, they produce similar analysis as the model. So I think it really is exciting that we can do that kind of work.

Rene Haas [24:26]
If you think about the next number of years, what are the things that are in the industry's way, not OpenAI’s way, in the industry's way. That said, gosh, if we had more blank - it's not you know, GPU is not a valid answer here - if we had more blank we would be able to go a lot faster.

Mark Chen [24:43]
Honestly, I think for now we need a lot more researchers and ideas. Right now is such a fruitful time to be doing research. I think we're limited just by the number of ideas we can come up with. Like you said before, if we could get the models to help automate that and help accelerate that, that would be phenomenal. But we need people to come up with ideas, implement ideas. But I do think another big thing is embodiment.

Rene Haas [25:12]
What do you mean by embodiment?

Mark Chen [25:13]
Right. Right. So we have a product called Operator. And what it does is it's a computer using AGI, meaning it takes as input computer screens and as output emits keyboard strokes or it emits click actions. And you can think of it as it's an interface to your digital work or your digital life, right? And when you think about the future, the extension of that is robotics, right? Like, essentially you're building an AI brain for a robot which can act in the real world. And I think there, you know, you’ll have bottlenecks around how quickly can we make the hardware work?

Rene Haas [25:51]
Do you think agents become everything over time? And we don't really have apps, we don't really have operating systems in the classic sense today in terms of you have to pull information out from a lot of different sources to get the answers. Do you think we get to a world where it's just an agent that exists somewhere?

Mark Chen [26:10]
Absolutely. I would love to have an agent where you just query for all of the information that you need. It should just be behind this layer, right, and you don't have to deal with it. I think Operator really does fulfill that promise if it works. And you ask, hey, you know, what, what is this information? It can check your Gmail and it can operate through any kind of interface you use in a computer.

Rene Haas [26:31]
And does it need to be a computer? In other words, can it be kind of anything moving - one of the movies I love to watch are the Back to the Future movies, which just in one sense, a lot of things are wrong, but a lot of things are rather clever. In Back to the Future 2, there's a scene where Marty McFly comes into his house and as he moves from room to room, the appliances in the room know the context. And they're either making his meal or they're turning on the news, or they're getting his boss on a video Zoom call, which, by the way, a video Zoom call that they were showing in the 1980s. Could you see a world where these agents are suddenly just now intelligent, running kind of everywhere, and back to your point, it's all kind of hidden underneath the hood?

Mark Chen [27:10]
I think they should be. Yeah. And you can imagine we have a model in the cloud, right? It's your OpenAI agent and all of these things are plugged into it. Right. And so they can understand the context as you go from one place to another.

Rene Haas [27:23] You are, you know, on a personal professional level, working at the hottest company in this space in one of the hottest times in this space. What is it like? Do you pinch yourself saying like, Oh my God, like I can't believe I'm actually in the middle of all this?

Mark Chen[27:35] Yeah, every day. I mean, it's a great privilege to work at OpenAI. You know, I started as a resident, actually at OpenAI. And what that means is someone who wasn't a Ph.D. in the field, right. And I'm grateful Ilya took a bet on me back then and he trained me up in machine learning. And, it's completely unpredictable, you know? And I'm just so grateful.

Rene Haas [27:58] Yeah. No, it's exciting to watch. And, you know, having been around Silicon Valley my whole career and watching these shifts from PCs to Internet to mobile phones, seeing what's going on here is, it's a thrill to be part of. Mark Chen, thank you for schlepping all the way down to San Jose from San Francisco and joining us today. I really enjoyed it.

Mark Chen [28:18]
Absolutely. Me, too. Thanks so much for having me.

Rene Haas [28:27]
Thanks for listening to this month's episode of Tech Unheard. We'll be back next month for another look behind the boardroom door. To be sure you don't miss new episodes, follow Tech Unheard wherever you get your podcasts. Until then, Tech Unheard is a custom podcast series from Arm and National Public Media. And I am Arm’s CEO, Rene Haas. Thanks for listening to Tech Unheard.

Rene Haas (00:07.298)
Welcome to Tech Unheard, the podcast that takes you behind the scenes of the most exciting developments in technology.
I'm Rene Haas, your host and CEO of Arm. Today, I'm joined by Alex Kendall, co-founder and CEO of Wayve. Wayve develops embodied AI for autonomous driving, and Alex has been in the driver's seat there since its beginnings at Cambridge University in 2017. All right, Alex, welcome.

Alex Kendall
Awesome, let's do this.

Rene Haas
So the story I tell a lot of people, the first time I had a demo of your technology was from Stansted Airport into central London. It was December in the UK, rainy, cold, not easy to see, two hours-ish, and we drove all the way in. It was assisted, but the technology was so amazing. I remember Masa was in the car with us, and it was so smooth that he fell asleep, which was quite a testament. But where has your technology come since that day? I think that was the end of 2023, maybe, maybe a year and a half ago.

Alex Kendall
Yeah, it was. I take that as a compliment that our AI could drive so smoothly that it's smooth enough for you to relax and even doze off for bit of the drive. So full compliment. So the approach that we've taken for autonomy is one where we treat this as an AI problem. Fundamentally, it's a really high dimensional decision-making problem, dealing with uncertainty, different signals, and having to put out a decision to drive a car in a way that has incredible generalization because the diversity of things you see when you drive on the road or in fact for any form of robotics is enormous. So eight years ago, we started off with the approach of end-to-end deep learning for driving and started off in a modest way with various different techniques of reinforcement learning and some Sim2Real approaches to learn everything from initial lane keeping to traffic lights, to roundabouts. But today, as I was mentioning, the strength of this foundation model is its ability to drive all around the world in many different types of vehicles with different sensor sets for different manufacturers and our ambition is to really see this launched as an embodied AI platform across a wide variety of different robots and work with the very best manufacturers and fleet operators.

Rene Haas
Definitely want to start with your upbringing and background, but I want to dig a little deeper in what you just said. In 2017, not obvious that using deep learning for automobiles or autonomous was an answer. What was behind your thinking that, yeah, this is going to be a great application for autonomous driving or anything autonomous?

Alex Kendall (02:36.032)
Ha ha.

Alex Kendall (04:58)
Both our company started in Cambridge. It was bit of Cambridge magic, but more seriously, I spent so much time when I was growing up building intelligent machines, playing with robots and just seeing how brittle and frustrating it is to build robots in a way that's the tang coded where you build the infrastructure rather than the intelligence. And that coupled with some years doing a PhD and surrounded by amazing opportunity to take a step back and look at where things were going. It was just very clear that end-to-end deep learning was slowly gobbling up the robotics stack through perception. And my PhD was able to do some of the early work that showed how to build different perception problems, figuring out where you are, what's around you, what's going to happen next with end-to-end learning. And it was clear that that was just going to keep going all the way through to control of the vehicle. And then when you think about what the future of robotics should look like, I don't imagine a car that's driving in a preset geo-fenced area following a map with a set of hand-coded rules. I imagine the future of robotics as intelligent machines we can trust to delegate tasks to them in a way that they can coexist in society, they can act and interact really broadly. And so for me, the way to build that is to actually develop a level of intelligence that can make its own decisions and operate safely within the environments that we live in. So it’s that belief coupled with a couple of like technical, you know, let's call them first steps that, that gave me the, the, the confidence to go, okay, let's go. We can go build this.

Rene Haas
So in 2017, what were the technological limiters relative to implementation? In the AI world, eight years ago is like 100 years ago. But if you go back to when you of conceived all this, what were the big technical barriers in terms of either compute capability and or sophistication of the models? Where were the big hurdles to climb?

Alex Kendall (04:29.144)
I think it's so important to work on things that aren't possible today, but are going to become possible and getting that strategy right on what's going to become possible at the right time. There's, of course, lots of aspects of luck at this, but also being prepared to take those opportunities when they become possible. And look, we've just seen DeepMind build the amazing AlphaGo, where there are more stakes in the game ago than there are atoms in the universe. There's an enormously large problem that was able to be solved through very efficient data sampling of the self-play algorithm. So there was that, but of course the game of Go is a very low dimensional space. It's a small checkers board, right? Whereas you think about driving, typical car might have, I don't know, anywhere from seven to 12 cameras, might have five radars around it. That dimensionality, you've got tens of millions of megapixels coming into the AI system. And so the dimensionality is enormous.
Some of the things that were really challenging then was how do you compress that information, understand it? Even how do you aggregate that amount of data? If you think about what's making self-driving possible, there's a ton of work that's gone on the hardware platform, the supply chain, access to data, and even more importantly, the off-board work and the data centers and the compute that actually allows you to train these kinds of models. And people often say that a system like GPT-4 was trained on maybe about a petabyte of data, but when we think about problems in robotics, I mean, today we're aggregating a data set that's over 100 petabytes. And so it's, it's really in another scale.

So my Dad had this, it’s funny, he wanted me to have that education and really be around sort of elite people in STEM, but he also wanted me to remain very Senegalese and so family, friends. So every vacation I did Air Africa, Air France, Paris, Dakar for maybe a week, a couple weeks, and that’s how I sort of stayed in touch with my original country and do to this day, this summer will be Dakar and London.

Rene Haas
When you say 100 petabytes that is on when you make that comment relative to robotics. 100 petabytes represents what in that world?

Alex Kendall (06:09.762)
Yeah, it's bringing together really diverse sources of data to be able to train this model because we want to make it so our AI can understand what various sensor sets can see. So we have everything from dash cams to surround camera, camera radar, camera radar, lidar data, or even internet scale text and video. And so it's all of that kind of diverse data. But of course, the multi-sensor data sets, camera, radar, lidar, of course, dominate in their size compared to the highly compressed dash cams.

Rene Haas
Well, one thing that I was struck by when we made that ride was not only the smoothness of the drive and how well it worked, particularly in central London, but there were not a lot of sensors and cameras on that car. You know, relative to what you see in the conventional AV1.0 with LiDAR and whatnot, I remember you telling me that you were not only achieving this kind of breakthrough using AV2.0, but with less sensors than a conventional LiDAR model or AV1.0. Talk to me about that a little bit and how that all comes together.

Alex Kendall
Yes, there's a couple of underlying principles that make our strategy possible. We talked about the first one, which is that taking an end-to-end learning approach, doing away with high definition maps, but taking an end-to-end learning approach. The second part of this is being flexible and agnostic to the hardware stack. More importantly, working with a lean stack that can be mass manufactured. So not the kind of hardware stack that you might retrofit at a small volume. But if you look at many of the today high-end vehicles, and in a couple of years, what will be mass market vehicles manufactured in tens of millions of unit volume a year.
These are vehicles that have hundreds of tops of GPU compute on them. They have surround cameras, surround radar. Some of them have a forward facing LiDAR, but working with a hardware stack that, you know, isn't like the robotaxis you see driving around Shanghai or San Francisco, but is like the vehicles that you see driving on the roads around the world. Those, that's the kind of hardware that we'd love to work with.

Alex Kendall (08:07.606)
And the benefit of that is that it enables you to bring this technology worldwide to really generalize it, to use cases that will eventually even go beyond passenger cars.

Rene Haas
But as to compute, know, so back to the conventional stack that exists inside of an unquote standard vehicle versus these look like spaceships, respectfully, kind of running around San Francisco. If you had more compute, is there something you look at and say, here's the minimum viable product in terms of this amount of compute I need the conventional vehicle versus if I had two or three times that number, call it 100 TOPs or 200 TOPs to baseline. Do you look at and say, I need that baseline and or gosh, if we had two or three X of that number?
The sky's the limit.

Alex Kendall
Yeah, I think onboard compute is certainly one of the limiting and boundary factors for what we're building or for any robotic system. So where we're at today, yeah, call it a 200-TOPs baseline. think this is certainly enough to get a really compelling hands-off driver assistance system that can drive globally in all kinds of scenarios. I think the jury's still out whether that kind of compute can get you to a general purpose eyes-off system in the near term. think there are certainly, it's amazing, like there are optimizations and approaches that just shave a factor of two or three off inference efficiency that can really be game changing or make or break for the system. But certainly, I think we can build such a system with some of the next generation chips that are going to be hundreds of TOPs, maybe a thousand TOPs. But there are a lot of, I think, very compelling reasons why we'll be able to compress these models down. Certainly if you narrow them to something like highway only or something like that, then you can achieve the level of safety on that sort of limited compute.

Alex Kendall (09:43.266)
When we think about AI scaling curves, which one of the most clear things for us is that they apply in robotics, just like we've seen them play out in large language models. In robotics, though, you don't just have pure AI scaling. You have system limitation issues. And inference compute's a clear one. Camera resolution or sensor fidelity's another one. And so you've got this dual battlefront where you've got to push AI scaling in terms of data and compute, but also work through systems issues, like empowering you with better and better actuation and vision.

Rene Haas
In autonomous is the limit for eyes off compute or sensor fidelity or combination of both.

Alex Kendall
Today it's intelligence. the vehicles that we're seeing that are coming into mass production, I think it's advantageous to have redundant sensing. So camera radar, for example, and the kind of sensor sets that we have right now. For example, if you were sitting with all the windows blacked out in the car and you only saw visualizations of the sensors that the car could see, you could drive at a level five or a general-purpose level.
It is possible for you to drive with those sensors with our human intelligence. We just need to be able to build the embodied intelligence to be able to do that. So I don't see a fundamental sensor limitation here today with where mass manufacturing is going. It's really, we've got to develop the AI and that's what I think we're on the cusp of launching.

Rene Haas
So, without giving away the secret sauce, tell us a little bit about how you develop the models. In other words, how you put out fleet vehicles, capture data, end quote, feed it into the model, and the model delivers some output relative to essentially assisted driving. So call it AV2.0 for dummies, if you will, in terms of how that all works.

Alex Kendall (11:22.552)
Yeah, it's a little bit like a striker in a football team. You know, the policy learning algorithms of actually how do you train the models gets all the intention, but actually is probably less than 1% of the work. 99% of the work is of course, all of the infrastructure to really make the iteration speed on these systems faster to develop. And then more importantly, how do you evaluate them? I think we're seeing this across all forms of AI, that actually evaluating and understanding and proving levels of safety in self-driving or more generally, understanding AI's performance is one of the harder challenges. But for the specific algorithms, I mean, it's a really powerful recipe of training a general-purpose foundation model that learns to understand many different physical tasks and many different physical domains. And then it's increasingly training and optimizing it for a specific deployment context, the specific SoC we're implementing into sensor architecture and the behavior that that manufacturer or fleet is after on their vehicles.

Alex Kendall
Both. So of course the classical machine learning answer is, okay, we collect a bunch of examples of that and go and train against it and shift and bias the behavior to resolve whatever behavior you're trying to do. Of course, there are many scenarios in driving that are either too rare or too unsafe to collect in the real world. And so we want to do that in simulation. I think one of the misunderstood things about synthetic data is that it replaces real world data. And I really don't see that being the case. I think if you have you know, more simplistic environments, then sure, you can hand code a simulator that can replace real data. But for robotics, the sheer scale and diversity that we operate in means that the very best simulators, and look, we put out one called Gaia, a generative world model, they're actually end-to-end data driven in themselves. Now the advantage is that they can learn to recombine data in new ways, and more I view them as giving you leverage on real world data rather than replacing it, because they need to be trained from the real-world data distribution.
But they can recombine and give you new experience of that data in a way that can help drive new reasoning and knowledge.

Rene Haas
Again, I tell the story many times of the Masa falling asleep in central London. But also my takeaway from that more than anything else was AV 2.0 is going to replace the current models. This is the future. And that's what I was thinking at that time. And if you look a year and a half later, there still are these kind of large vehicles running around with 1.0 systems. What's it going to take for 2.0 to just wipe the 1.0 stuff off the face of the earth? Is it just time?

Alex Kendall
Time's certainly going to help. I mean, it's been really interesting to see the just seismic shift in the industry over the last year. Eight years ago when we started working on this, this was like deeply contrarian and dismissed by almost everyone in the industry. Some of the proof points we've had coupled with, of course, just the sheer broad breakthroughs that the AI field has had. And I have really materially shifted this in the last couple of years. And even to the point where two years ago, I couldn't even get conversation with most automotive manufacturers to now that we've got the CEOs of these companies stepping out of our car and those that haven't fallen asleep with comfort going, wow, I want that in my product yesterday. And now it's a case of, okay, how do we engineer this into these products? How do we go and validate the levels of safety that are required? I think we're going to very quickly see, at scale, hands-off point-to-point driving. I mean, we're seeing some products out there now and what our solution will do is enable many different automakers to launch that kind of capability. But the real inflection point in safety and value comes when you make it eyes off. so I think collectively we're in a huge challenge and I wouldn't necessarily call it a race. I'd call it yeah, mean, ambition to bring that product together.

Rene Haas
Going back to when you started the company. What was the vision for starting Wayve? What was your purpose?

Alex Kendall
I mean, our mission statement is about reimagining mobility with embodied intelligence. And so it's all about building AI systems that enable us to have autonomous machines. Now, when you look at all the different forms of autonomy from healthcare, manufacturing, domestic robotics, I think what was very clear to me was that in automotive and in autonomous vehicles, this would happen far beyond these other spaces because there are so many more factors and conditions that are mature there. There's global supply chain, there's the availability of data, there's a compelling business case, there's proactive regulation in place. All of these aspects are still very nascent in these other spaces. So I have no doubt that they will all become huge game-changing opportunities as well. And I would love to see our AI generalize into them as well. But it was clear that starting in autonomous vehicles was the place to start.

Rene Haas
Did you start it when you finished your PhD work? Where were you when you started it?

Alex Kendall(16:36.17)
I was simultaneously writing up my PhD thesis and raising a seed round of funding

Rene Haas
That's courageous.

Alex Kendall
and trying to find some local lab mates to join me on this wild journey and turn down their seven figure big tech offers and build a prototype in the garage there with me. it was a pretty fun couple of months.

Rene Haas (16:57.28)
And why did you not take the seven figure big company offer versus rubbing two sticks together on your own in a small little barn house in Cambridge? I always find when I'm speaking to founders like yourself who have been unbelievably successful, that the courage it takes to sort of start that kind of thing when you kind of know the opportunity is big, but you've got behemoths with lots of capital and lots of people competing against it. What was the fork in the road that had you choose to found a company?

Alex Kendall
Yeah, was a very, very bizarre time. You can imagine coming out of a $10,000 a year PhD stipend. We also, not only that, had offers from a couple of big companies to buy out the technology we're working on. I think it was really interesting, actually. Some of those companies that wanted to buy out the technology wanted to use it to drive perception or data labeling systems. Even there, I was just very skeptical because, of course, the point of a computer vision system is to make decisions. And why stop there? I think, you know, it was very, very clear you needed to go to end-to-end learning for control, but it really just came down to my core values. driven by learning and adventure, and I'd almost reached some of the limits of what was possible in the academic setting I was in, in terms of needing to build a team and had the amazing opportunity to come to Silicon Valley and spend some months with a robotic startup there, Skydio, that was doing incredible work, but learn what it was like to be part of a venture-backed mission-driven team.

Rene Haas
Were you advising them or what was the role that you were?

Alex Kendall (11:59)
Yeah, was an intern. I implemented the very first deep learning system for them. Gotcha. I got the drone to go from following people wearing a blue t-shirt to just following people in general, and a very small part of what is an incredible company today. But I guess that exposure opened my eyes to, what's possible with venture capital? Yeah. But moreover, it was a means to fuel the passion that I had, which was building this technology.

Rene Haas (18:50.072)
So you went from an intern while you were obviously getting your postgraduate work, but now you're the CEO of a tech company, but you haven't had years and years of management training and such and whatnot. What were some of the things that you've learned about yourself and learned about building a company in those early days? I think folks would love to hear the story about what you learned as a founder.

Alex Kendall (12:29)
Every moment you have is, it's interesting how it can just be an opportunity for learning. Cause one of the interesting things I observed was, look, there's, there's PhDs and there's PhDs. It's very, very different experiences you can have. But some of the things I remember, you know, going through that experience was, I believe you have to let your results do the talking and first have something of substance. But if you are fortunate enough to have something of substance, how to communicate it, how to talk about it, how to give speeches and tell a narrative that can drive further progress was just such an important skill. I turn up to conferences and share work where five different people would have done that work and those that could communicate it, was a real advantage. I remember even just the importance of being able to describe and share a narrative that connects people and aligns them around a vision. But then even just in the early days hacking and building together these prototypes, particularly when there's nothing that can path find for you. There's no existence proof that it's possible. How exciting I found that was, and how every little breakthrough you make, just how much energy it gives you and then how every setback would just also just drive so much determination to go solve it. But I remember the day that we had the very first reinforcement learning system that could learn to actually lane follow and drive down the road. And I remember going out, we were so frustrated after weeks and weeks of work that one weekend I just went in to the house, got the car, went to the test site myself and just, you know, thrashed it out to the point where it worked. And I remember coming back to the team with that result and the delight and moreover how limited it was and how much more there was to do was a really magic moment.

Rene Haas (20:51.448)
You also did something pretty tough in that you started a company 2017, doing very physical things, obviously, in terms of autonomous. It's not obvious in 2017 how this is going to be used. Pandemic hits in 2020. Did COVID, was that a setback for you guys in terms of your development? Or how was that period like for Wayve?

Alex Kendall
That was, it was interesting how much we problem solved to work through that. For example, bringing cars to people's houses so they could test in the local area around the block where they were and still actually get time driving, sitting in the car and actually feeling the system drive. But then also, man, it was a tough time personally, cause I was, I was back in New Zealand for a lot of it back home.

Rene Haas
My gosh.

Alex Kendall
And, of course the time zone difference between we were largely a UK company there, but, the time zone differences exactly 12 hours between London and New Zealand. So I was waking up at midnight, working till midday ish and, think getting, getting some sleep and I was back there for some personal reasons, but yeah, that was, that was a rough nine months.

Rene Haas
Did it slow you guys down in terms of if you look back and say, gosh, if we didn't have the pandemic, i.e. access to roads, butts in seats, or you guys just muscled through it?

Alex Kendall
We muscled through it, but counterfactual if the pandemic wasn't there, I'm sure it did have a slowdown effect. I think when you're building any kind of robotic system, being there physically, being able to experience and brainstorm and whiteboard is just so important. So finding ways to do that virtually required some creativity. But of course, I think we saw this effect that it did bring teams together.

Rene Haas
So you guys have about 500 folks, right? 550. And I've worked in small companies where when we got to 180 or so, that was a bit of a tipping point relative to everybody knows everybody and they know what people are doing. For you guys, how's it been scaling to that number and has it felt different as you know, or at this size and growing?

Alex Kendall (14:58)
Yeah. And even, you know, we've, as a team, we've just gone through, I think another one of those inflection points. It's just amazing to me how, I spend so much of my time designing the, and thinking about the machine of how to operate such a team. And I think that's the most complex thing about building the company more so than the technical strategy. But as soon as you get good at that process, you outgrow it. And it's just been my repeated experience growing to where you are today, but this latest one that we've gone through is taking us from more of a centralized roadmap to one where it's empowered teams that have all the cross-functional skills to go execute against an outcome with the right communication structures to keep them all aligned because building this kind of product requires so many different areas to come together, but we've just gone through a change where we've tried to reduce some of the coupling and empower our teams to move faster.
Of course, collaboration does come at the expense of bandwidth and it's important to make room for that, but to make room for it in a deliberate way where it's needed. And you see many of those differences from when we were a 50 person team and everyone knew everything to even now getting comfortable with not knowing everything. I certainly don't have the visibility we had when we were 50 people or even 20 people. And so being able to grow and empower structures to run where you're deliberate about where you collaborate and where you trust and empower has been one of the biggest things I've seen here.

Rene Haas (24:16.27)
How does feel for you personally? I mean, there's so much written about, you know, end quote founder mode, and leaders need to be in the details. And I know at my company, 8,000 people, obviously I'm not in every detail, but I like to be in the details around the stuff that I think is utterly critical to the success of the company. For you, I mean, you're managing an incredibly complex product and roadmap, technically. How much are you in the details?How does feel for you personally? I mean, there's so much written about, you know, end quote founder mode, and leaders need to be in the details. And I know at my company, 8,000 people, obviously I'm not in every detail, but I like to be in the details around the stuff that I think is utterly critical to the success of the company. For you, I mean, you're managing an incredibly complex product and roadmap, technically. How much are you in the details?

Alex Kendall
I think highly and I think I pride myself that although I took one of our friends, Jan Lacoon for a demo drive the other day and he continued to debate with me really, really deep down into algorithmic details and I was embarrassed that it got to a point where I couldn't go any deeper, which would not have been me a couple of years ago. But in general, no, I think I'm an exceptionally detailed orientated person and actually have had to, to coach myself to try and step out and empower a team in many ways. But I think one of the things that I do spend a lot of time on is trying to fly at a lot of different levels, to be able to make sure that there is a quick, line of communication throughout the company. But then, you know, when it comes to actually, aggregating a lot of that signal and, figuring out where there is red tape or problems to be cut or changed to do so in a way that empowers people to drive and those outcomes.

Rene Haas
One analogy I've always heard, which I like for CEOs, is that you need to behave like a helicopter. In other words, be able to see above everything, but at a moment's notice, dive down to three inches above the problem.

Alex Kendall (25:49.002)
But I mean, you guys have gone through an amazing last year as well. What's, what's been the biggest, biggest thing for you that you've felt?

Rene Haas
You're turning into interviewer mode here.

Alex Kendall
Why not? Got take the chance.

Rene Haas (16:19)
I think scaling a company through transformation is a big challenge. We're going through a lot of changes in terms of moving to more of a platform-based approach, which ends up being a more complex solution. And as a result, you're going to be bringing in new muscles into the company that you haven't brought in. At the same time, you go through change. And thankfully or unthankfully, depending how you look at it, many companies go through change during a time of crisis. In other words, their business is in peril and it's either change or die, which can accelerate people's acceptance of change. We've been doing very well. Arm’s had a great history. The last number years have been terrific. So when you're trying to instill that level of change in the midst of a successful business model, that can be a challenge from a leadership standpoint. But that's probably the single biggest one. It's a great time as we're chatting about this technology.
If you're on the cutting edge of more and more compute and complex technology and complex algorithms, what more to ask for. You mentioned Jan, who is a great guy. I love talking to him as well. And now you've sparked something in your comment that I'd be curious to get your viewpoint on. He's been very vocal, and I've chatted with him also about this, that the work that people should be spending on vis-a-vis LLMs, cetera, is not how to make the LLMs more interesting, but what's beyond them, right? What's kind of beyond the transformer-based model from an algorithmic standpoint, and there's no way I'll go anywhere close to as deep as you guys can do. When you think about that, what is kind of beyond that? Because right now, in a good way, in terms of more and more compute, people are just adding more and more gigawatts, more and more data centers. The more pounding they can do on the problem, the more reinforcement learning, the more compute, the better it is. But is there a next level paradigm shift relative to, from an algorithmic standpoint, that says, hey, when we get to this threshold, everything is different?

Alex Kendall
Yeah, I think there is absolutely a cascading set of S curves of new approaches to bring in. It's been really interesting to see, I mean, broadly speaking, think it'd be fair to comment in the large language model and cognitive AI space that shift from being a science and engineering problem more to a product problem of how do you exploit new data sets and integrate into different product workflows. But one of the interesting things in embodied AI is, is yeah, we absolutely are seeing that new ideas are needed because there's still that big safety bar.
You need to get and not only get, but measure in a way that has quite, quite unbounded and the domain that you can drive in. So I share a lot of Jan's views that I think absolutely for this next S curve, this next, next jump, we're going to need to find new methods that look, there's a lot of detailed work that goes into whether you're going from, you know, whatever flavor of learning structure from convolution nets to transformers, to RNNs, LSTMs, all the, all the different, you know, architectural structures that will continue to make incremental gains. But the representation learning methods for me are the ones that are really driving more general-purpose understanding and in particular how we can get more effective learning signals out. Cause I think if you can predict and understand a scene, then actually acting and engaging within it should be, should be very efficient.

Rene Haas
Anyway, as we wrap up, it's been fascinating, Alex, and what you guys have done has just been amazing. You're a New Zealander, I'm an American, but we're both running Brit-based companies. You've had a lot of help from the UK government in terms of just clearing regulatory hurdles in terms of getting your stuff on the road. How do governments need to continue to help in this? Are governments a regulatory, one of the bigger hurdles to get through? When you think about kind of eyes off driving?

Alex Kendall (29:36.268)
Look, I think there's a lot of misunderstanding with regulation because I actually feel very optimistic about it in autonomous driving. And in general, think there is a hard question about general purpose AI regulation. But when you think about specific AI applications, whether it's medicine or education or specifically for driving, there is a fantastic set of regulation as well as understanding of the risks and the opportunities in that space. So I think actually in autonomous driving, you know, we are today seeing very proactive regulation, even though the technology is not widespread. We helped the UK government write into law last year, the automated vehicles act 2024, which legalizes autonomous driving. We work a lot at the UN-level to drive global harmonized legislation. And I know the US is starting to do a lot more there too. So in general, I think for applications like autonomous driving, we're actually in quite a healthy spot. And yes, we need to do more.
We need to implement it more quickly, but I think progress is compelling. And for me, it's an important problem, but the long pole of seeing this technology launch to scale still remains the intelligence, the AI, the science.

Rene Haas
So last point, plug for the audience. what countries can we find Wayve technology these days?

Alex Kendall (31:17.464)
Rene, we're all around the world now. We're in the UK. We've got a fleet, some of our cars being tested in Germany, the US and Japan now. We do road trips all around Europe. We were in Italy last week. We were in Canada a few weeks ago, but that's just our dev fleet at very small scale. What I'm excited for is to see this launched into mass market, consumer and commercial vehicles around the world. And we're working with a number of automakers and fleets to make this possible.
Two were publicly announced, namely Nissan and Uber. We're incredibly excited about. hopefully in a car that maybe you own in a couple of years, or maybe you’ll be driving and that'll be there soon.

Rene Haas
And that car will have Arm inside, so we both win. Alex, thanks so much.

Rene Haas
If you want to learn more about Wayve and their software strategy, keep an eye out for an upcoming conversation with their VP of Engineering, Silvius Rus, on the Arm Viewpoints podcast. Thanks for listening to this month's episode of Tech Unheard. We'll be back next month for another look behind the boardroom door. To be sure you don't miss new episodes, follow Tech Unheard wherever you get your podcasts. Until then, Tech Unheard is a custom podcast series from Arm and National Public Media. And I'm Arm CEO, Rene Haas. Thanks for listening to Tech Unheard.

Rene Haas (03:04)
Aisha, thank you so much for joining.

Aicha Evans (03:21)
My pleasure. I’m so looking forward to this. We’re going to have fun.

Rene Haas (03:37)
So Aisha, you and I had lunch a while back.
You told me a little bit about your background, which was utterly fascinating in terms of not only where you grew up and got started, but how you got into engineering in this field. So recast that for the audience, it’s just a fascinating story.

Aicha Evans (04:22)
Yeah, I do remember that lunch too and many dinners since. Look, I’m from Senegal, West Africa. My dad was in telecommunication engineering back then, sort of France was wiring up Senegal and other countries. And I grew up, I ended up growing up in Paris back and forth. And every time I was in Senegal and I wanted to talk to my friends in Paris, I couldn’t really. I remember this is pre-internet and what have you. And so we had a rotary phone. And yeah, I decided to figure out how to hack it because it had a little lock.

Rene Haas (04:57)
Ha ha.

Aicha Evans (04:58)
Lock on it so you couldn’t make phone calls. Now in retrospect, really, it was sort of silly because then the phone bill came and my dad was like, what is going on here? But I just saw the difference when you have technology available, what it makes possible versus when you don’t in two countries that I love. And then I happen to be good at math, physics, not biology, by the way. And so that’s how I fell in love with STEM and building things, hacking things, and I do like the combination of hardware and software. And then eventually came to the US to study computer engineering and the rest is history.

Rene Haas (05:37)
So born in Paris and then lived in Senegal or born in… So, and what was the back and forth to Paris then? Why that?

Aicha Evans (05:41)
The other way around. Born in Senegal and then yeah.

So my Dad had this, it’s funny, he wanted me to have that education and really be around sort of elite people in STEM, but he also wanted me to remain very Senegalese and so family, friends. So every vacation I did Air Africa, Air France, Paris, Dakar for maybe a week, a couple weeks, and that’s how I sort of stayed in touch with my original country and do to this day, this summer will be Dakar and London.

Rene Haas (06:20)
Amazing, amazing. And the lock on the rotary phone, was that to disable country codes? What was that lock about?

Aicha Evans (06:28)
It was, it turns out to first of all, back then not a lot of households had those. And so there were issues with sort of people going to somebody else’s house and making phone calls. And so this was locking it so that only authorized people with the key could make a phone call because it was so expensive except that one of the unauthorized people was me, the daughter who wanted to talk to my friends in Paris.

Rene Haas (06:44)
My goodness gracious.

Fantastic. I grew up in the United States, but my mother was Portuguese, my father was German who grew up in Lisbon. similar thing in that they would make calls overseas. And we didn’t have a lock on the phone, but I would remember usually every month when the phone bill came in, my father being in complete stupor, finding out what calls my mom had made overseas.

Aicha Evans (07:22)
It was crazy back then. I tell the kids today who at the moment noticed are on a video call with friends all around the world and they’re like, you just don’t know how good you have it.

Rene Haas (07:30)
Yeah. Yeah, no, the world got unbelievably flat. So where did you go to university?

Aicha Evans (07:39)
GW in DC, which is yet another story. Finally, he agreed, my dad, that I could go to the US, but he wanted me to go in DC where he had a lot of friends at World Bank, IMF. There was also a small contingent of Senegalese students because he was like, okay, in Paris, I lost control of this child. She’s this little French girl. She’s smoking, dating, drinking, going out. We’re not doing that again.

Rene Haas (08:02)
You

Aicha Evans (08:07)
And the funny story is very shortly after being in the States, I met my American husband. And so the recipe didn’t really work out too well.

Rene Haas (08:18)
And studied engineering or computer science? was your studies?

Aicha Evans (08:22)
It was computer engineering, so I had this thing that I didn’t like electrical engineering. I like hardware. I like to build things. But you could already tell that computers were taking over. And computer science I liked, but I was going to miss the hardware. So there was this mixed degree that’s kind of nice where you do a little bit of both, which worked out really well for me.

Rene Haas (08:41)
Interesting, interesting. I didn’t know that story, but my degree is also in computer engineering and actually for the very same reason. Electroengineering, electromagnetic fields and waves, that stuff I did, power circuitry, transformers, good to learn. But computer engineering, when I was in college was really the time of, the PCs were just starting.

Aicha Evans (08:57)
You get it.

Rene Haas (09:07)
And I had worked, around with Commodore 64s and VIC-20s in high school. So very similar. Computer engineering was exactly what I was attracted to as well. So I knew we had something in common here.

Aicha Evans (09:20)
There we go, now it makes sense.

Rene Haas (09:22)
And then what? So you’re out of GW. What is your first role?

Aicha Evans (09:26)
I am at, well first was stay for grad school and don’t stay for grad school but married, my husband had a job in Austin and so that was kind of like, you want me to move where? Austin, Texas, are you crazy? But we did and it turned out to be a good decision so I went to work for Sky, well back then it was called Brook Three which was, yep.

Rene Haas (09:49)
Oh sure, yeah,

Aicha Evans (09:51)
A big startup and…

Rene Haas (09:52)
Yep.

Aicha Evans (09:52)
Totally bombed the interview too. I don’t know why they hired me or at least that’s how it felt. But it was great. I was the youngest by many, many years and they just took me under their wing and worked on essentially AV sync. Back then recording live video was just starting and streaming and then quickly, I think four years or five years in with my big mouth and always having opinions moved into management. At first the tech lead and then from there went to, where did I go? Skyworks and started working on cell phones. I was just talking this morning about sending those first text messages on a small little LCD screen and saying, yeah, it’s gonna be computers in your pocket soon.

Rene Haas (10:32)
My goodness. So very, but we’re two for two here, because my first job out of college was also in Texas, and I was with TI, and so gosh, at Houston, so we’re two for two here. But Brooktree, I remember they were doing like RAM DAX and analog, how big was Brooktree? Because when I joined TI, you know, a monstrous company, how big was Brooktree?

Aicha Evans (10:59)
About 500 people working on the project was with, I think it was EchoStar maybe, and basically what were you gonna do with storage and replay and synchronization, the audio and the video are recorded separately, so they sent me to Sarnoff Research Labs in New Jersey to learn about all that, and that was my area of the chip and the product.

Rene Haas (11:08)
Sure, yeah, yeah, yeah. So when I joined TI, they were a giant company. I think they were the largest semiconductor company in the world. To be honest, I knew no different about joining a small company versus a big company. But Brooktree must have been interesting. Did you have a desire to join a company of 500 people? Because when you’re going to go of college and you’re new to everything, it’s not obvious how the culture is going to be at a small company versus big company, how one’s going to fit. did you end up there? How did you end up at Brooktree?

Aicha Evans (11:59)
Okay, so I have to be careful here. So this is what happened. I joined a big company. And I spent two weeks at the big company because it was the right thing to do and because I was on a senior project at GW with them. But then when I went to work, I was like, this is not going to work. This dude,

Rene Haas (12:08)
Okay.

Ha ha ha.

Aicha Evans (12:29)
Was like, you’re gonna do this? I felt like I had a roadmap of tasks with no thinking for five years ahead of me. And so I quit. I was like, and they were like, how, two weeks. So then they said, hey, are you sure you’re intimidated? I said, no, this is not gonna work. I have opinions, I need to think, I need to be part of things, and I can just tell I’m not gonna be happy here.

Rene Haas (12:29)
Wow. Within two months? Two weeks.

Aicha Evans (12:57)
And so they gave me two more weeks of vacation to kind of get my mind right because they thought I was just scared and I never went back. And during those two weeks, I looked at the Austin Statesman, think was the thing, the Sunday pages. And there was this ad from Brook Three that said, come make waves with us together. And I called them and I said, hi, I’m Aisha Evans.

Rene Haas (13:18)
Gosh.

Aicha Evans (13:22)
I’d love to work for you and they’re like send your resume in I sent my resume in and then they called me in for an interview and I interviewed and then I got an offer and I was screaming first time I had ever done anything by myself for myself

Rene Haas (13:30)
Gosh. But you had obviously shown some level of entrepreneurial risk taking and no, really seriously, and courage because to know something at that young age isn’t right for you and to make a decision that fast is pretty amazing. that, looking back, is that something you look back and say, gosh, what was I thinking? I was crazy? Or do you look back and say, that was me and that’s how I think about things.

Aicha Evans (14:04)
No, I mean, I’m not saying it’s good or bad. It’s just me. By the way, there are also other sides of the issue where I love to cook. This is something that makes me happy, cooking and making dish and doing dishes. And so you are now in the US. Even though DC is very international, there is no Senegalese food and I need my Senegalese food. So I make my Senegalese food. My friends tell me how much they love my Senegalese food. I come into a little bit of money and I decide it’s a good idea to start a restaurant. So I buy a restaurant with my American boyfriend who we’ve been together for three months. And I learned a lot about leadership at that restaurant. Turns out the dishwasher dude, he’s the boss. And so I did that for a year now, like back to computers, please. Let’s go back to the regular program. So sometimes it’s kind of crazy. In retrospect, I want to ask my parents-in-law.

Rene Haas (14:35)
My gosh. Ha

Aicha Evans (14:58)
I mean, they were probably like, this woman, what is she doing? But I don’t know. That’s just me. I’m weird.

Rene Haas (15:03)
Was the Senegalese restaurant done in parallel with your career or was it a hiatus and you went up and said I’m going to go do a restaurant for a bit?

Aicha Evans (15:11)
So first I thought it was going to be done in parallel. It quickly became clear that that was not going to happen. So I ended up taking a semester off and then going back to school.

Rene Haas (15:23)
Unbelievable, I did not know that story, that is amazing. So after Skyworks, is that what brought you to Intel or was there something in between there?

Aicha Evans (15:34)
It was being married 10 years, not wanting to have children because anytime I’m around children, it’s loud, it’s complicated, and you have a responsibility. And then all of a sudden, you decide that, no, no, no, you want to have children. And so it was, OK, I need to slow down and have kids. It ended up being difficult also having kids. Now I can talk about it publicly because it’s been a long time, but had a very difficult miscarriage.

Rene Haas (16:02)
Mm.

Aicha Evans (16:02)
And so I was like, okay, the new project is to have a kid. So I left. But then in the meantime, until after I had my first child, was like, hey, we’re looking for wireless people because of WiMAX. And we happen to live in Portland, Oregon. So it was very practical. So I kind of, you’re going to say this woman, I kind of eventually said, fine, I’ll come to work for you. But

Rene Haas (16:19)
Sure.

Aicha Evans (16:30)
For the first year, no travel because I had a three-month-old brand new baby that I really wanted to have. And I think I said something like nine to four. And they said, okay, fine, no problem, just for the first year. And of course, I went in there and, you know, that’s not what ended up happening. I ended up working like crazy and traveling. But I also figured out how to, I wouldn’t say balance because I think that’s a myth, but integrate being a mom and working. So yeah, it was Wimax that got me there.

Rene Haas (16:43)
Right. So do you have, and I have two daughters and their mom did not work when the kids were growing up, which I think was hugely beneficial to them. And now my youngest daughter has had a baby and she’s also staying home for a bit. But I look at that and then I think about your story, which is just amazing. i mean, what…

Aicha Evans (17:15)
Congrats!

Rene Haas (17:28)
How did you do that?

Aicha Evans (17:42)
You’re too kind. So a few things. So first of all, you know how in when you’re on an airplane they say you have to put your oxygen mask first before you help anybody else? I like working. I enjoy it. It makes me happy.

Rene Haas (18:00)
Yeah.

Aicha Evans (18:07)
That journey from the young Senegalese girl all the way to here required a lot of managing myself and understanding what’s important to me. So I like working, I like creating things, I like having an impact. I love my kids, I have two of them, one is now 19, the other one 17. I have a wonderful husband, a wonderful support system. By the time, remember, I said I waited 10 years, so how it worked out for me is that by the time I had them, I could make choices and afford choices in terms of what I did myself versus what I got support for versus frankly what I outsourced. And so I built a wonderful flexible support system around myself and I have a wonderful husband too and we partnered in doing that. And along the years sometimes I sped up from a career or impact standpoint. Sometimes I took a step back and just kind of ebbed and flowed.

Rene Haas (19:01)
For folks listening who might be thinking about those choices, is there anything that you look back and say, had I known X, I may have done something a little bit differently.

Aicha Evans (19:19)
Not particularly when it comes to raising the kids. would say, well, the jurors start out, they are young, so we’ll see. But so far they are wonderful kids. Look, the thing I would have told myself is something that applies to almost everything in my life. That same entrepreneurial, risky, seeks sort of crazy things, wants to have high impact, sometimes also goes out of control and is too intense. And so I would say, Take a chill pill is what I would have told my younger self. In general, it works out. then, yeah, that’s probably what I would have told myself. As far as the advice, you just need to be consequential. Be happy and be consequential. So when you make the decisions, make sure you understand the pros and cons, and be happy with the pros and live with the cons.

Rene Haas (19:51)
That is absolutely wonderful advice. now you’re at Intel, certainly not a small company at the time, and probably pretty close to the peak of their strength as a company. And you were there, what, 12 years? what were the areas that you worked inside of Intel?

Aicha Evans (20:27)
Indeed. Twelve years. So it was mostly 10 years of wireless because when I came, when I went in with Wimax, I was kind of like, wow, they’re either onto something incredible or something’s really wrong. So it turned out that the idea was correct, but the implementation, mean, Wimax is a great technology, but when you look at the incumbency of 3GPP back then, that was almost mission impossible in terms of the ecosystem, network operators. I ended up meeting, it was sort of fortitude, I was in some open forum type thing with David and Sean. David Perlmuter back then ran all of engineering and Sean ran kind of the business. And they were sort of outside the CEO, they were kind of like running the company. And I put my hand up and I said, hi, I’m Aisha Evans. I have a question, I work on WiMAX, and I see all these challenges, and so I’d like to understand what’s the goal? What are we trying to do? And so they responded, but that got me an audience with them afterwards. And so it ended up being that Grove had seen that the PC and the connected world were going to converge. And so having a wireless portfolio was important. License, IE modems, unlicense, IE Wi-Fi, Bluetooth, GNSS, that type of stuff. I basically worked on building those products for mostly PCs, a little bit of edge devices, and then a smartphone from a company that I’m not allowed to name.

And that’s what I did for a long time, enjoyed it, had a great team worldwide and was in the soup with operators and base stations and it was really fun. Back then MWC in Barcelona in February was a must. And yeah, that’s what I was doing. But in doing that, I started, it almost felt like a startup inside of Intel. I don’t know how to say it because everybody else was doing processor memory and what have you.

Rene Haas (22:27)
Absolutely. Mm-hmm. Yep.

Aicha Evans (22:40)
And then there were changes and then I received the directive to make a couple of changes. One was a processor change and the other one was to build internally, i.e. not at TSMC. And that’s when I encountered the beast.

Rene Haas (22:49)
Yeah, and we promise not to go too deep on the Intel thing, but the one thing I always, I look at Intel and I think a super power to some extent is vertical integration, I thought to some extent because I go back to my early days at TI and then back in the day, we built everything internally. We built testers and memory access and all the programming. And it just seems to me that

Aicha Evans (23:02)
It’s okay.

Indeed.

Rene Haas (23:26)
The strong coupling between the process and the product is a pretty amazing power if they can get it right. Without getting too deep in the Intel broader strategy, when you think about vertical integration in semiconductors versus the horizontal model, do you have a perspective on, in the long game, which should win, or one way the other?

Aicha Evans (23:47)
I don’t think about them that way. First of all, I always kind of go back to what I call the helicopter view, meaning above the forest. Success is poison. If you don’t manage your success and you lose track of what is the basis of your success and is that basis a truth that is self-evident and should continue or be modified, thou shall run into trouble. And so there is absolutely no doubt that the vertical integration is a strength, but if it’s not done in a way that is flexible and modular and can quickly adapt to trend, either in the building, the material, the testing, or the cadence of the market, trouble arises. And then you go into protect and defend mode, and that is never, ever a winning strategy.

Rene Haas (24:41)
I 1,000 % agree. i mean, the technology graveyard of tech companies is filled with companies that have done just that. And I was in Austin doing a fireside chat at South by Southwest. And we were literally chatting about that and that just how punishing our industry can be if you miss a window and you think your past success is a predictor of a future strategy, you are going to be in huge trouble.

Aicha Evans (24:53)
Yeah.

Rene Haas (25:07)
You made a very brave move to go to a technology area that I think everyone would know and would disagree that autonomous vehicles and autonomy in general is what’s going to happen. So your move to Zoox, now you’ve gone from something that’s very big to something that’s not so big, but in some ways was fairly well capitalized. What made that jump for, what was the jump to Zoox all about?

Aicha Evans (25:36)
Well, towards the, kind of, I don’t think you should wake up every morning thinking 100 % of the time you’re gonna be happy at work. But you know, the majority of the time you should be happy and energized and you’re learning and you know, and you’re contributing. And towards the end, at Intel, I could feel that, you know, I had more unhappy days than happy days. Let me put it that way. Not because people were mean or anything like that. It’s just, I didn’t agree. And so I’m starting now to, you know, I’m very curious. That’s how we’ve met, but I’m stuck. I’m one of those who says, hi, my name is Aisha Evans. I need some help, please. I don’t understand. May you please help me? And so I had met a lot of people in the industry and a lot of folks were telling me, should go do this. You should go do that. And…

Rene Haas (26:11)
Yep, absolutely.

Aicha Evans (26:26)
I don’t know that I knew what I wanted to do exactly, but I knew how I wanted to spend my time. And so I was getting a lot of offers and I’m not trying to sound arrogant or anything like that. That’s just what happened. A lot of phone calls. And eventually I got one about Zoox. By the way, I first told the recruiter, I don’t think so, because we had, Intel had bought Mobileye. So I had a little bit of a window into that ADAS AV world.

Rene Haas (26:50)
Yeah.

Aicha Evans (26:55)
and we were also supplying wireless cards to EV folks. And so I was like, these people, no, this sounds crazy. But then a couple of people that I really trust told me, hey, you should go talk to them. They might be onto something. And so first, of course, I meet board members. I like them. And then I meet Jesse Levinson, one of the two co-founders. We walked around Zoox for about a couple hours. And I don’t know, it’s one of those when you know, you know.

I just fell in love with the idea and the product. I felt the vibe. I could see myself spending the time that way. It’s a product that matters. It’s been a while since there’s been, in physical transportation, since there’s been like a big revolution. And it just makes sense. I mean, I was like, we’re not building that many more roads and how many more cars can we stuff? Plus I hope that 10 years, 20 years from now, we ask ourselves how it was okay for 40,000 plus people to pass and that’s just okay. And so it just, they had a business model that made sense to me. And so I came home and I told my husband, I think I’m leaving and this is what I’m doing. And so then we went on vacation for I think two weeks in Jamaica and I sat there on the beach and.

Rene Haas (28:11)
Hahaha

Aicha Evans (28:19)
You know, I trust my intuition a lot and my intuition didn’t say don’t do this. So I came back and told Intel and then took like a little bit of detox time and then started in February 2019. Haven’t looked back and I still giggle every day.

Rene Haas (28:35)
Six years, and the progress that you’ve made has been incredible. I do want to touch on that, but I want to dig into a comment you just made because I see that in your personality. You’re not afraid to ask for help. I’ve been a CEO now for three years. You’ve been one for six. It’s not always easy for the CEO to stand up and say, I need help. People look at us and say, the buck stops there. You’re the answer man or the answer woman. How do you balance that because you’re great at it in terms of the I’ll ask for help when I need it But at the same time, I’m the CEO and should I be seen as asking for help?

Aicha Evans (29:17)
So I think being open and transparent is extremely important. i think telling people the why before you get to the how is really important. My job right now at least is to get some 2,800 people or so from many different backgrounds, Some are traditional automotive, some are new automotive, some are AI geniuses, some are mechanics, some are embedded people, some are policy and regulatory people, some are comms and marketing people, some are product experience. I am not an expert at all that. Number one. Number two, I already know what I think. What’s most important is what I don’t think and that’s good. And so my job is to treat everybody as a customer. kind of take all of that input in, synthesize it. If there are better ideas, say thank you and tell me more and let’s do it. If my idea is bad, I’m in a safety critical business. People trust me with their lives in our robotaxi. I would want a culture where if people think we’re doing something wrong or unsafe that they tell us. So to me, it’s more like earning it.

Rene Haas (30:33)
Yeah, perfect, perfect answer. That’s how I feel about it. And in fact, the CEO probably has the good excuse that can say, I’m an expert of nothing, so teach me something. And I find that my own job here. I learn every day from folks who I interact with. So much has happened in Autonomous in the short time that you’ve been with Zoox, although six years maybe doesn’t seem like a short time. When you started, large language models, vision-based learning, that was not a thing. And now, six years into this, what has changed, if anything at all, relative to how you think about autonomous and the rules and how that gets handled in the light of these new models that seem to be just incredible in terms of learning?

Aicha Evans (31:30)
First of all, AV is a space where it’s important to ask a question first, which is, is there a driver behind the driver’s seat? Yes or no? So if our view is that if there’s a driver behind the driver’s seat, but that driver can be distracted, that driver is not fully in charge of the driving, they are just in charge of making sure they are available to take over. basically supervised versus unsupervised. So I think that LLMs have accelerated things, but for us, where there are no manual controls and there is no human driver, so there is no takeover because we don’t want the customer involved in the driving, LLMs have been more helpful in simulation, in the magnitude and the speed at which we can process edge cases, the understanding cost, just because you have to, in our case, supplement cameras, that doesn’t mean you have to, like, if we were redesigning our robotaxi today, I’m pretty confident it would have less sensors than it does today. And then in terms of also anything that requires a lot of estimation and forecasting, so it’s been an accelerant for us. We are not yet at the point where we’re thinking,

Rene Haas (32:37)
Right.

Aicha Evans (32:51)
Without any manual control, it will do the driving by itself. We think that from a safety standpoint, and the easiest way, not everybody lives in this technology, but if you prompt one of, pick your favorite agent, and you give it a prompt and it gives you an answer that’s slightly right and a little wrong, that’s not a big deal. Well, when it’s in real time control of a robot taxi, that is a big deal. So that’s the gap we’re maneuvering. But we think that over the next, I’m sure that gap will narrow and narrow and narrow over time given how fast things are going.

Rene Haas (33:25)
Yeah, that does feel like something that time can address and the models are moving remarkably fast in terms of not only their learning and reasoning capability, but just simply you throw more compute at it. To get to a point where you think the models are actually end quote good enough to be a good augment to what you’re a couple years away, multiple years away,

Aicha Evans (33:55)
There are already a good augment now. There are many times when you have perception and mapping and localization, but then really the pipe is perception, and well, prediction, planning and control. First of all, prediction and planning and control have basically merged. And second of all, every trajectory has essentially, there is a machine-learned version that is generated. It’s just we don’t pick it all the time. So it’s happening.

Rene Haas (34:22)
Right. to get to a point where it’s picked the majority of the time or 90 % of the time, how far away is that?

Aicha Evans (34:31)
I think 90 % of the time is very soon. The problem is when we get to 95, 99 % of the time, I think to get to that last 1 % is where it’s going to be very hard. Because even if it can, we want to know that it can every single time. And that’s the little funnel that’s difficult to deal with. But it’ll come.

Rene Haas (34:35)
And well. is the threshold, you mentioned 40,000, 50,000 that death a year in auto accidents, which is just awful. Yet it seems like with autonomy, the threshold for accidents is,

Rene Haas (35:26)
Essentially zero.. How, how, where do you think the threshold level is relative to the tolerance for, for accidents? And I know this is kind of a crazy subject relative to trading off humanity, but in reality,

Aicha Evans (35:27)
Mm-hmm.

Rene Haas (35:54)
Does it need to get to essentially zero for this to be able to be broadly and widely accepted? Is it regulatory? Is it insurance? How do you think about all that?

Aicha Evans (36:02)
No, so I’m trying to give you the best answer that I can give you. I’ll answer it in a philosophical way. I think you have to be trusted. People have to trust that you know how important this is and you’re trying for zero. And there has to be a track record holistically, whether you’re dealing with regulatory, the community, your own engineering, your own methodology. There has to be a track record. You have to develop a track record of transparency, communication, and how you talk and how you do things so that people trust that you’re trying for zero. I think if that, and also it has to happen. mean, look. the human benchmarks are fairly clear. We can argue about 10 or 20K here and there, it’s around, no matter how you look at it, it’s around 10 to, let’s say, 50K when it comes to collision. It’s around 30 to 100 million when it comes to, sorry, when it comes to injuries and then fatalities, right? So, You can’t really, you know, if, for example, if Zoox deploys a robot taxi and has something really bad happen after two miles, we do not deserve to be trusted. So I think you have to build all those muscles and have a lot of transparency and make sure that you build your safety case in a way that some of the tougher incidents happen down the road so that you have enough of a track record of that so people know you tried.

Rene Haas (37:21)
Right, right, right.

Aicha Evans (37:42)
And then they come and then you have to explain what happened and make sure that you, nobody has the impression that you didn’t tell them what happened. Even if you did, know, perception is reality, you’re a CEO, you know how that works. And then you have to show what you’re gonna do differently.

Rene Haas (37:57)
Totally. Do you think it’s a couple years of having no accidents, for example? I was wondering, because I think about air travel, which is unbelievably safe, even given the issues that have taken place horribly over the last number of months. Statistically, it’s incredibly safe. And statistics will show, and have shown, that the vast majority of airline crashes was pilot error. Yet when I look at autonomous, just to your numbers, I would almost think that the math will be there very soon, that they will be less than what you just cited. Yet I also feel like today’s culture and philosophy is almost a zero tolerance. And I just wonder how we get through that threshold.

Aicha Evans (38:45)
So it’s interesting you should say that. I work in foster city, so I see the planes landing all the time. Aviation, no matter what we say, over a lot of miles in the air, over a lot of takeoffs and landings in a very ubiquitous way, has shown and proven that it tries for zero, very rarely.

Very seldomly, when something bad happens, but we have this contract that we trust the process. And I think that autonomy, especially unsupervised autonomy, is working its way there. But the bar should be high, because we’re trading people’s lives, and that’s important.

Rene Haas (39:31)
Yeah, 100%.

Aicha Evans (39:34)
One last thing though, You have to deliver value too. Because you know what? You can decide that airplanes are not safe and you’re not flying. Well, that’s going to limit a lot of things in your. And we have to, so we have to deliver value at the same time.

Rene Haas (39:38)
For sure.

Yeah, Just a last couple of questions, Aisha. When I became a CEO, a lot of folks asked me afterwards, what surprised you about being CEO? And there were so many things. But a small one was that I didn’t quite realize was just how the eyes are on you all the time. And that’s from an employee standpoint. used to, the job I had before running the company was I had the number two job in the company, which was a big job, et cetera, et cetera. But I would notice I could walk down the hall and people would maybe not look at me and not say hello. And when you’re the CEO, I think everyone is watching you and making sure that not only they wanna say hi or hello, which is obviously great, but they’re also reading your body language and such.

Aicha Evans (40:41)
For sure.

Rene Haas (40:49)
You’ve been at CEO for six years. What was something that has surprised you about the role?

Aicha Evans (40:55)
Definitely what you said, but in addition, how bloody lonely it is, It’s just lonely. It’s very isolating. It’s funny because you’re isolated in public. On the one hand, you have to do exactly what you said, shoulders up, because I’m on stage, people are reading my slightest body language or smile or nod of the head or how I’m talking to somebody or wagging my finger.

Rene Haas (40:59)
Haha

Aicha Evans (41:22)
And at the same time, yeah, sometimes I just need to let loose and I’m not sure or I’m scared or I am hilariously happy. And it’s just, it feels like you’re just, you’re always in public, but you’re also so lonely.

Rene Haas (41:36)
That is very well stated and that’s why I do this podcast and I have people like you on because it’s my one time of the day to feel not such a lonely person. Last question and this is something I’ve never quite been able to figure out the answer to. Why is the bread in France so much better than anything you can find anywhere else in the world? What is the secret sauce there?

Aicha Evans (41:56)
Look, I don’t know and I’m not a baker, but hang on, hang on, hang on. But I have something. one of my daughter’s best friends, he’s French and the mom is French and they’ve known each other since middle school. And I can tell you that I’m very mad that they just moved to Paris because when we are invited, when we used to be invited, to get an invite to their house for dinner.

Rene Haas (41:58)
What?

Aicha Evans (42:24)
In Los Altos, I’d be like, my god, this is going to be like the bread in Paris. And so I finally asked her and I said, hey, how come you’re able to do this? And I’m not able to buy this anywhere. Now you’re sitting down. She imported every single ingredient from the butter to the flour to the salt. mean, every single ingredient she imported to make the bread.

Rene Haas (42:38)
I’m sitting down.

I have to believe that’s probably the case because what I have found is either one of two things. People have imported all the ingredients or they do what you just did and don’t tell me what the answer is in the beginning for somebody who lived in Paris. Aisha Evans, thank you so much for spending the time with us. This was wonderful.

Aicha Evans (43:10)
My pleasure, Rene. Good to see you.

Rene Haas (43:12)
Merci beaucoup.