Arm Blueprint checks in occasionally with futurist Matt Griffin, founder of the 311 Institute, to get his take on compelling technology trends in a future built on Arm. He’s talked about the future of digital retail and health care applications with former Arm Fellow Rob Aitken and more recently the metaverse with Arm’s Remy Pottier (you can listen to part 1 here and part 2 here). Blueprint chatted with Griffin recently to get a sense for how he sees emerging AI-based tools – conversational AI, virtual and augmented reality, and the metaverse – are starting to affect workflows. And he talks about how his 10-year-old son produced a book using ChatGPT and other AI tools.
Blueprint: Let’s start with augmented reality and talk about some of what you’ve shared with the world in your technology Codex. There are some issues with some of the devices, particularly the headsets, which we still haven’t gotten over. The nausea problem due to latency issues, for instance. How do you see that evolving in the next few years because clearly people are trying to commercialize it?
Matt: Low latency is crucial for the human brain to perceive a virtual environment as real. A small delay of a few tenths of a second can be noticeable to the human brain, causing you to feel off balance. There are several ways to achieve this. 5G NR technology reduces latency to sub-millisecond speeds. Although we are still some way off from achieving this goal, progress is being made. Companies such as Meta and Samsung are developing new codecs to improve the compression of video streams, which are notoriously difficult to compress.
So companies like Google, as well as Samsung, are using adaptive codecs to try to compress this video stream that we are seeing here on our call, this kind of 4K video stream by typically about 10 to 20%. In compression terms, that’s absolutely fantastic.
There’s also a new technology, called foveated rendering. Foveated rendering is a technique where the AI only sends 20% of the data to the computer, allowing the brain to fill in the rest of the image. This approach has been shown to reduce data transmission by 80% while maintaining resolution in virtual reality.
Another issue with VR headsets is their bulkiness and size. A trial in the U.S. asked people to work in virtual reality for a day, over the course of a few weeks. Pretty much everybody was coming out of that basically feeling horrible. The people were just coming out going, “there’s no way I could ever work in virtual reality for, decent chunks of time.” So when you actually have a look at the development of what we call meta optics. Meta optics is the study of creating new optical devices, such as “meta lenses”. These lenses are only a few atoms thick, made from thin polymers. They are being developed for use in glasses.
The development of virtual reality glasses, helped by advances in communication, codec, and AI technologies, is improving the user experience. This results in higher-resolution virtual reality worlds. The use of photorealistic game engines like Unreal Engine 5 is driving this progress, with developments maturing over the past 6 months to a year.
Blueprint: OK, ChatGPT: Transformative technology or fox in the henhouse? Your thoughts.
Matt: Right. ChatGPT will affect about a billion jobs and about $10 trillion worth of GDP. You get a sense for the potential impact when you consider that my 10-year-old son used ChatGPT and then MidJourney to write and illustrate a professional runner’s guide in under a day. I wanted to show him the technology, but on the other hand, I treated it like an experiment: If you gave these powerful multibillion-dollar artificial intelligences to a 10-year-old, what would they do with it? It took 20 minutes to write the book. It took about six hours to do the illustrations. It then took the two of us about eight hours to proofread and format it in things like Adobe InDesign. We started selling that last week, and the proceeds go to charity.
Blueprint: How did he do the writing part of it? Did he just ask a series of prompts?
Matt: So firstly, Caiden had to have an idea. He loves running. So basically, he wanted to write a running book or, more accurately, generate a running book. He went into ChatGPT basically, and simply created a prompt saying, “create a list of chapters for a book titled ‘coaching for performance runners’” and started the chat.
GPT then generated a list of about 12 chapters, such as “introduction to new nutrition,” that kind of stuff. Then the prompts that we put in basically were “now for each chapter heading right up to 1300 words of content.” GPT would go off and it would write generally up to about 1,000 words of content. We then did some tweaking of the prompts, so we said, right now, put in examples, don’t turn it into a list – these sorts of things. So we tweaked it and did some formatting. Once we were happy with the result, which, and this is all within 20 minutes, we then copied it, dumped it into a Word file, and then I took that word file, formatted the text, put it into InDesign.
We chose our sort of book format and layout. And then once we’d done that, did some sort of tweaking. We then went to MidJourney, which is a text-to-image artificial intelligence tool, and then Caiden started putting prompts into that. And in that case, we had to do a lot of tweaking, but if you have a look at the book, you’ll see the images that we actually came up with.
Blueprint: What did you find interacting with MidJourney?
Matt: What we did find with MidJourney, though was that they’re very difficult to tune. So say for example, the images that we created, if I wanted the person who was set in the center of the image to be to the left, you’d put a prompt into MidJourney saying, “now move the person who’s in the middle to the left of the image,” it just wouldn’t do it. And this is the reason for using this – to see what their upper limits were, see where they broke. There were biases, but if you watch the video, what I do is I go, what we created, how we created it the impact basically of these technologies but then also our findings as well.
So for example, ChatGPT was quite generalistic. And it was trying to make up some stuff. If you wanted ChatGPT to write something in like an upbeat style, it could do it, but it was bad at that, and then MidJourney basically had a habit of turning images into a little bit dystopian look.
But for me to generate the images that MidJourney created, I estimate that I would probably have had to have paid someone about £10-15,000 to produce the images. I would’ve had to manage them. I’d would had to have give them prompts and I’d have probably had to have spent about two weeks to four weeks, backwards and forwards working with them on the images.
With MidJourney, I spent £20. Because I wanted it to process the images fast because it couldn’t really be bothered to wait. And we generated thousands, and we did it all literally in hours. So when you talk about the future of work, one of the things that we typically talk about is we talk about is how organizations want to scale efficiency. As we get bigger as an organization, you want to do more things more efficiently, faster, and cheaper. If Caiden and I had decided to write the book that you see there the old-fashioned way – like the way that we would’ve in 2022 – I reckon it would’ve taken about two months. The images would’ve taken about up to another month. And so it would’ve taken us about 2-3 months to produce that book, and it would’ve cost us, say, £10,000, which I’d never have spent. But that entire book was produced in under, really, almost under a working day. And it was produced for £20.
Blueprint: So I read this morning that ChatGPT CEO Sam Altman said it’s either going to be the most transformative thing to hit humanity ever, or it’s “lights out for all of us.”
Matt: So, I had a chat with the U.K. government recently and we talked about the pros and cons of certain tools. As you work in the content industry, I think it’s important to know the pros of these tools. As these tools become more integrated, smarter, and sophisticated, they will help you do more in less time, like writing that book you’re working on.
ChatGPT, just think about it from a creative perspective: This technology will impact over 235 million creatives in a $6 trillion industry. It also affects coders, people in academia and more, but that’s a different topic. With these tools, access to certain skills will become more democratized. For example, I could never have created the images in that book without these tools, and neither could Caiden. I’m not an artist, but now I almost have an artist on staff. If I didn’t know how to write a book, I now have access to a ghost writer on demand. I could create a recipe book now using these.
Here’s the rub: So on the one hand, these technologies, as I’ve been saying for years, democratize access to skills for everybody. Now that means I can do new things faster, which is great for efficiency, but the artist that I would’ve employed that I no longer need to employ, that’s bad news for him or her.
Creating content becomes increasingly democratized, and on demand, whether it is videos, adverts, or social media content, because we are tying ChatGPT up with Canva, and you can automate your entire, pipeline of social media messages that you want to throw out as we see the democratization of content.
But it also means that any content that any of us creates is going to be much more difficult for people to find. Because if I can create a book in a day, what happens when everyone else can create a book in a day? Say for example, I’m doing a book on running, all of a sudden, the market today might have, what, 10 million books on running.
But now I’ve got a really bad marketing problem because I’ve got a running book but so does everyone else. You’re either going to be a destination for content, where people always come to Matt’s site because I’ve got fantastic content, or I figure out how you actually use all of these different algorithms effectively to try to boost my content over the other 10 billion running books. So increasingly trying to get your content viewed – a content tsunami.
And that’s before AI even starts producing its own content. Because I also say that, at the moment, Caiden had the idea and then we created the book together. But I could just tweak an algorithm to automatically get the AI to create a book every two seconds.
Blueprint: Moving on: Crypto is in crisis, however blockchain has huge upside. Where do you see that at the application level?
Matt: Frankly, crypto isn’t dead. The future of money is not about money, but rather the future of value. For instance, value can be traded without using U.S. dollars. Blockchain has moved beyond its hype cycle and is now being implemented and used to build things. Decentralized apps like Jack Dorsey’s new social network, Bluesky – in the context of Web3 – are also emerging. Blockchain applications are being developed and deployed globally. As a general-purpose technology, blockchain has the potential to transform various fields, including computing, storage, file sharing, and organizational structures.
It will take about 10 to 15 years for many things to become blockchain-based. The growth of blockchain and decentralized apps is increasing, but a lot of it is hidden since blockchain is like plumbing in a house: It is essential, but often not noticed until it needs to be repaired. Blockchain is a phenomenal technology, but it is like an engineering technology, hidden in the fabric of systems.
Blueprint: What about security?
Matt: The gas pipeline hack and the SolarWinds hack are examples of the need for improved cybersecurity. Blockchain technology has the potential to address this need through its secure and decentralized systems. For example, a sovereign blockchain system can secure the authenticity of software updates and prevent similar hacks to SolarWinds.
Blockchain also has numerous applications including NFT marketplaces, crypto, Web3, decentralized autonomous organizations, secure communications, distributed computing, and even combining with AI to decentralize access to powerful AI tools. Blockchain technology can also be used for authentication purposes, such as images, artwork, newsfeeds, etc.
Blueprint: Thanks, Matt.