EP 46: Emad Mostaque (Founder/CEO Stability.ai) - On The Future of Generative AI, Creating Real-Time Movies, Societal Impact of AI

In the 46th episode, Stability AI CEO Emad Mostaque discusses the future of AI. From being only years out from generating entire movies in real-time to how it’s revolutionizing the healthcare industry and could help identify cancer. Plus its benefits in education, charities, the potential regulations, and societal impact, and Stable Diffusion becoming the most popular open-source software in the world. (0:00) Intro (1:03) Introducing Emad Mostaque (1:27) Generative AI (9:52) Computing power and AI (12:23) How Emad got into AI (14:49) Open-source AI (20:19) Growth of Stable Diffusion (21:57) Lensa (25:39) Power of Creativity (30:00) Ramifications of AI (34:09) State of the industry (37:32) Business model for AI (43:28) AI use cases (48:11) Societal impact (51:39) Becoming a public figure (55:41) Outro Mixed and edited: Justin Hrabovsky Produced: Andrew Nadeau and Rashad Assir Executive Producer: Josh Machiz Music: Griff Lawson 🎙 Listen to the show Apple Podcasts: https://podcasts.apple.com/us/podcast/three-cartoon-avatars/id1606770839 Spotify: https://open.spotify.com/show/5WqBqDb4br3LlyVrdqOYYb?si=3076e6c1b5c94d63&nd=1 Google Podcasts: https://podcasts.google.com/feed/aHR0cHM6Ly9mZWVkcy5zaW1wbGVjYXN0LmNvbS9zb0hJZkhWbg 🎥 Subscribe on YouTube: https://www.youtube.com/channel/UCugS0jD5IAdoqzjaNYzns7w?sub_confirmation=1 Follow on Socials 📸 Instagram - https://instagram.com/cartoonavatars 🐦 Twitter - https://twitter.com/cartoonavatars 🎬 Clips on TikTok - https://www.tiktok.com/@cartoonavatars About the Show A Saturday morning podcast hosted by Logan Bartlett (Partner and Managing Director at Redpoint Ventures) covering the tech news with his friends and other people with industry expertise. Executive Producer: Rashad Assir Producer: Leah Clapper Mixing and editing: Justin Hrabovsky Check out Unsupervised Learning, Redpoint's AI Podcast: https://www.youtube.com/@UCUl-s_Vp-Kkk_XVyDylNwLA 🎥 Subscribe on YouTube: https://www.youtube.com/channel/UCugS0jD5IAdoqzjaNYzns7w?sub_confirmation=1 Follow on Socials 📸 Instagram - https://www.instagram.com/theloganbartlettshow 📱 X - https://twitter.com/loganbartshow 🎬 Clips on TikTok - https://www.tiktok.com/@theloganbartlettshow About the Show Logan Bartlett is a Software Investor at Redpoint Ventures - a Silicon Valley-based VC with $6B AUM and investments in Snowflake, DraftKings, Twilio, and Netflix. In each episode of The Logan Bartlett Show, we sit down with the people behind today’s most important startups and extract the tactics, lessons, and frameworks they’ve learned the hard way. Conversations span hiring to GTM, product, growth, fundraising and everything in between - collectively forming the ultimate playbook to make you a better CEO, investor or board member. Tap follow and enable notifications to stay ahead of the game.

Speaker names marked ? are inferred from speaking time and the show’s known hosts — they are not voice-verified.

• 0:05Logan Bartlett?

  Welcome to the 46th episode of Cartoon Avatars. I am your host, Logan Bartlett. Welcome back for break. Thanks, everyone, for bearing with us as we took a pause over the last couple of weeks. We're excited for this episode. This, what you're going to hear on this episode is a conversation that I had with Ahmad Mustak. Ahmad is the founder and CEO of Stable Civility AI, which is the largest contributor to stable diffusion. Stable diffusion is the fastest growing open source project of all time. It's one of the leading platforms in generative AI. And Ahmad and I had a really interesting conversation about a bunch of different things, but we dive into the state of artificial intelligence

• 0:46Logan Bartlett?

  today, why this is possible, when it wasn't in the past, where this is going in the future, how he differentiates versus competitors like OpenAI. Really fun conversation and appreciate him for powering through. He was a little sick as we were doing this. So it was, it was a fun conversation and I appreciate him, him doing it with me. And so before you hear that, we talked a little bit about this before break, but we are going to make a more concerted effort to get people to like and subscribe and share and review the podcast itself. And so if you're, whatever platform you're listening on, if it's YouTube, if it's Spotify, if it's Apple, whatever it is, if people could go ahead and like and subscribe and leave a

• 1:31Logan Bartlett?

  review, share with a friend, all of that stuff. We're trying to figure out exactly what direction to take this in. And so that validation and feedback and also the growth that comes along with all that stuff is, is super appreciated. It's not something we had been comfortable asking for to date, but as we kind of figure out what direction we're going to go, we'd love to see more shares, more reviews, more views, more likes, all that stuff. So really appreciate everyone's support in doing that. And so without further delay, what you're going to hear now is the conversation with me and Ahmad Mustak from Civility AI. All right, Ahmad Mustak.

• 2:09Logan Bartlett?

  Did I say that right?

• 2:10Speaker 01

  Yep.

• 2:11Logan Bartlett?

  Perfect. Thank you for, uh, thank you for doing this. Founder of Stability AI, one of the, uh, main contributors to stable diffusion. Thank you for, uh, for coming on here today.

• 2:23Speaker 01

  It's a pleasure, Logan. Nice to have you here.

• 2:25Logan Bartlett?

  Yeah, totally. So, um, maybe at its highest level, we can start off with what, what is generative AI? How would you define that for the average person?

• 2:34Speaker 01

  So I think, um, everyone has had a kind of the concepts of big data because the whole of the internet previously was on big data, large, large models built by Google and Facebook and others, uh, to basically target you ads. Ads were the main part of that. Um, and these models extended. So it had a generalized model of what a person was like, and then your specific interests, you know, like Ahmad likes green hoodies or, you know, Logan likes black jumpers, um, that then extended the previous to what the next thing was. They're like extension models inferring what was there. Generative models are a bit different in that they learn principles from structured and unstructured data. And then they can generate new things based on those principles.

• 3:17Speaker 01

  So you could ask it to write an essay about bubble sort or a sonnet about Shakespeare or the teenage witted ninja turtles and it can do that. Or in the case of some of the work that we're most famous for, you enter in a labradoodle with a hat in a stained glass window and it understands that and then creates that in a few seconds. So I'd say that's probably the biggest difference between this new type of generative AI and then that old type of AI. So the way that I always say is we've moved from a big data era to more a big model era. Cause these models are very difficult to create, train, uh, which is why only a few companies such as ours do it.

• 3:53Logan Bartlett?

  And the key point there is the, the predictive nature of these models and the ability to actually take not just what was given to it, but also act on all the other things to sort of self-create in that way. Or what was the distinction that you were drawing there?

• 4:11Speaker 01

  Yeah. So like if you built a dog classifier previously and then a new type of dog species came along.

• 4:20Speaker 01

  Then the classifier wouldn't be able to understand it because it didn't really understand the concept of a dog.

• 4:25Logan Bartlett?

  It was just responding to all the history of things that it had been fed in terms of, I guess fed is an interesting analogy for a dog here, but all the history of data, they didn't have any understanding that it was a dog. It was just, hey, here's all the parameters around which this thing seems to find.

• 4:42Speaker 01

  Is dog-like, yeah. You know, this is one of the issues with self-driving cars. Like you have this whole world of things that you're used to, but then what happens if something happens that is not on the training set, right? So in 2017, there was a breakthrough in what's known as deep learning, whether this is paper attention is all you need about how to get AI to pay attention to the important things as opposed to just everything. So we moved from just analyzing everything to analyzing the important things. And this is what's led to a lot of the transformative breakthroughs that has allowed AI to get in very narrow areas to human levels of performance in writing, reading, playing Go, playing Starcraft, all sorts of things, protein folding, et cetera as well, and other breakthroughs that are not human, as it were.

• 5:24Logan Bartlett?

  And so it feels, it feels to me and probably the average person that this all came out of nowhere, but we've had sort of incremental progress in AI over the course of the last 20 years or so. Can you, can you give a quick primer on like back to the days of Deep Blue and, and chess and Deep Mind and Go? Like what have been the, in your mind, the historical points along the way that sort of led to this avalanche that feels like have just, has just happened?

• 5:55Speaker 01

  Yes, like machine learning was kind of classical paradigm. Actually, one way to think about it as well is that you've got two parts of your brain, the part of your brain that jumps to conclusions and the logical part. So the world as it is, and that's holy crap, there's a tiger in the bush, right? So classical AI was the more logical kind of way, and it was based on more and more and more data, again, big data. So when Deep Blue beat Garry Kasparov, it's because it could think more moves ahead of him. It just did pure crunching. Of the numbers, it looked at every chess match and then outperformed him. And eventually people knew it would get to that point, but they didn't think it happened quite so quickly.

• 6:29Logan Bartlett?

  And that was, that was in nineteen ninety six, ninety seven ish. And basically chess is kind of a constrained game in a lot of ways. Like there's only so many moves that can be made. And so you can computationally run through the history of all of the moves and figure out what the next best action is.

• 6:44Speaker 01

  And then also look at all of the previous moves. So Garry Kasparov, I believe, could think five, six moves ahead, but then Deep Blue could think seven, eight moves ahead. And it was just giant supercomputer, literally, you know, that kind of beat him. And I was like, okay, that's the case. Um, then humans started playing with computers and computers started playing with humans. And now the best chess players are humans working with computers, which is very interesting. But it was again, this very defined space. By contrast, Go was um, a game that people thought couldn't be beaten by this mechanism. Because a Go board, Chinese chess, has too many computational possibilities. So you can't think X moves ahead because just exponentially more compute required.

• 7:29Speaker 01

  Um, so DeepMind Research Lab out of London, now owned by Google, built a system called AlphaGo that created a computer that again learned principles and actually played against itself. So they didn't even look at historical games for the later versions. They pitched it against Lee Sedol, who was, I think he was seventh or ninth dan. He was basically like the Magnus Carlsen of Go, so far ahead of everyone. And everyone's like, ah, there's no way they can beat him. It drew with him once and beat him like another seven times. And I was like, wait, what? Um, like I said, without doing that massive levels of number crunching, because it learned what was important in moves and the principles to do moves.

• 8:07Logan Bartlett?

  That was twenty sixteen when this, when this happened.

• 8:10Speaker 01

  Exactly. That was kind of with the reinforcement self-supervised learning, which was one component of this, before the deep learning came, which were transformer-based attention learning, which was twenty seventeen, which is the next step above that. So there's a few things all happening at the same time, along with an exponential, again, I'm a mathematician, so don't use exponentials lightly, literally, a lot of these things look like exponentials because they are exponentials, increasing compute availability.

• 8:35Speaker 01

  So what happened there is that then it was very interesting. The uh since that point where everyone's like, holy crap, he's got beaten. He's his level has gone up in go, but then so is everyone else. So if you look at the average level of go players, it's been like this for about three decades. And now it does that. Because a computer could think in brand new ways and think about new principle both ways to do it. But now humans computers got even better. Then there was the transformer based architecture paper. Um I'm skipping over a lot. There's a lot of stuff happening in deep learning. And it was this attention based system whereby it paid attention to the most important parts of a given data set. That led to breakthroughs like GPT three in twenty twenty. Uh GPT three is a model by OpenAI, which is a research lab, primarily backed by Microsoft,

• 9:20Speaker 01

  focused on artificial general intelligence. So how do you make an AI that can do just about anything? Um that could write like a human. So you give it Legolas and Gimli and it'll write you a whole story in the style of Lord of the Rings. But what it does is basically it guessed what the next word in a sentence is from a giant corpus of text, actually not a big, a few terabytes, a few thousand gigabytes, that was then run on a gigantic supercomputer. So supercomputers kind of had a linear increase in their capabilities over the years. And you know, you see things like the Apollo landing is like same compute as your iPhone, right? But that was still quite linear. Over the last few years, um led by kind of Nvidia and these GPU moments,

• 10:05Speaker 01

  you've had an exponential increase in supercomputer. And these models lent themselves to, you take a relatively small amount of data, uh like text writings of the whole of archive or PubMed or a scrape of the internet or like a billion images with uh captions. And then you put it into the supercomputer and the supercomputer looks at the connections between the words and the images or the words in a sentence and how they line up to figure out what should come next. So this was a big breakthrough in that you didn't actually have to build a custom algorithm for everything anymore. There was one set of algorithms, like you had to do a good level of customization. But the key edge and key differential was no longer how big is your data set?

• 10:52Speaker 01

  You know, or seeing how customers use the data set, it was just how much compute do you have? So more and more compute was applied to these models and then this broke through and they got bigger and bigger and bigger. So like GPT three was one hundred and sixty seven billion parameter model. That's the kind of, you say it's the kind of things that it knows.

• 11:13Speaker 01

  And then it got to five hundred billion and then bigger and bigger. These large language models, as they were called, that could be human level in answering questions. You know, and this technology started to proliferate because when you get to human level, it was great. But it didn't proliferate that fast because it was slow. And it was expensive and it required a lot of technical expertise to even run these models, let alone create them. And again, like the super compute levels are just beyond belief. And I'd get to that. At the start of last year, on the image side, which is one of the areas that we're focused.

• 11:49Speaker 01

  The OpenAI released something interesting called Clip, which was an image to text model. So you could also generate text descriptions of images. There were some generative models before that. And so you had a generated model and then you had a model that could tell you what a generation was. And so a bunch of groups came together and said, what if you bounce them off each other? Well, that tells you what an image is and one tells you how to generate an image. You could converge to better images and that's what kicked off this whole image revolution.

• 12:18Logan Bartlett?

  Your conversion is from, from language, so it's just text based or speech based or whatever into images. You're bringing two different modalities together.

• 12:28Speaker 01

  Yeah, so the two models bounced off each other. So it'd be like a dog in a stained glass window and it produced a version and then the image to text model would be like, eh, that's not that good. It looks like that, but then there was the other prompt and it made adjustments. Then it went back and forth, back and forth, back and forth. So you got something that looked a little bit like a dog in a stained glass window. And then teams around the world, led by a lot of people at Stability and OpenAI and Meta and some other places, started to think, how can we really crack through this? And it just went a bit crazy to get to the point now that you can generate a photorealistic image of anything in about a second.

• 13:06Speaker 01

  And again, this is part of the exponentials, like the amount of compute that we're using now as a private company that's 14 months old is 10 times the compute of NASA put together, or 10 times the compute of the fastest supercomputer in the UK. It would have been the fastest supercomputer in the world just six years ago. So for a private company to be able to access that is a bit insane.

• 13:27Logan Bartlett?

  And how did you, you have a little of an unusual past into this, or path into this. So, so can you talk through like how you actually ended up at the forefront of a lot of these things?

• 13:39Speaker 01

  So yeah, I was quite lucky through life. I was a hedge fund manager. I was a video game investor. I took a break when my son was diagnosed with autism. And I realized that AI could be used to try and solve some of these things. This is the old school AI. Because for people who know about autism, autism spectrum disorder, there's no official treatment or cure, or nobody knows actually what causes it. And so I was like, what if we did a literature analysis of all the different things that people think could cause it and try and figure out some commonalities with AI? And then identified some things in the brain, GABA and glutamate balance. So GABA calms you down and glutamate makes you excited. So when you take Valium, your GABA goes up.

• 14:20Speaker 01

  When your brain is too excited, it's like when you're tapping your leg and you can't focus and pay attention to things, right? And so kids with ASD are often like that in that they can't pay attention to form links between words and images and concepts. Actually, very similar to these diffusion based image models. So a cup can mean cup your hands, a cup that you've got like that, a world cup, you know, maybe Argentina or France will win it, who knows, recording just before that. So you need to calm down the brain somewhere. But when I looked at it, like there were 18 different things that led to that potentially. And certain treatments to make some kids worse, some kids better. So we did a lot of drug repurposing on the end of one. And then I was advising governments and things at the time about AI and all sorts of other topics.

• 15:02Speaker 01

  I was like, this is really powerful technology, but you know, I'm not a doctor, so I'll do my best to tell other people about it, but it's okay. But then about a few years ago, I realized that actually this technology could change the world. So we used it first in education in a small set and refugee camps around the world, the charity my co-founder runs, Imagine Worldwide. But that's going massive and there'll be announcements next year. And then working on the United Nations AI response on COVID-19 as well, because again, it's this thing where it's multi systemic condition, no one knew what was causing it and that knowledge needed to be organized. Had loads of bureaucracy through that, lots of companies promising stuff that wouldn't deliver, really got into this sector and realized this AI is possibly the most powerful thing we've ever seen because human level means a lot, right?

• 15:46Speaker 01

  And the only people that could build that are was basically the big tech companies plus OpenAI and a couple of others. And none of them wants to release it open because it is powerful and powerful means also dangerous, right? There's always an upside and downside. I don't believe technology is neutral. But the way they were doing it, it would never be released. So it would only be available to a select few and those select few could create any image in seconds or write an entire story. When would it ever go to India or Africa or places like that? So that's why I thought this is infrastructure just as important as the internet for the next step in kind of human ability and it should be open source. And then also it's a better business model as well, putting on my hedge fund manager hat.

• 16:28Speaker 01

  Because all of our vital infrastructure for the internet, servers, databases, DevOps, it's all turned open source now. And the simple business model is scale and service. People come to you when they want to scale it, well, customer versions of it. And I thought that's the winning business strategy. So that's how I set up Stability as a mission-based organization with a profit-based focus. But the profit is making these models available to everyone and customizing them and scaling them for everyone as well. It has been interesting as, you know, not a Silicon Valley native or anything like that, and talking to people about why we should let people have access to this technology and people are generally good, not bad. But yeah, you know, it's been quite a ride.

• 17:07Logan Bartlett?

  I want to play that back to you a little bit because it's the how you actually got into it is is such an interesting part of this. So you were you were working at a hedge fund and you took a break because of your son's autism disorder and you were able to do a bunch of different stuff with traditional AI to kind of figure out how to make sense of the different drugs and the causes and all that stuff. And then Did you see some, was it, was it OpenAI and what they were working on? And you saw that? And how did actual stable diffusion and the involvement and all of that stuff come to be? Because it sounds like you were around the industry, but did you meet someone and say, or was someone already working on stable diffusion, the open source project?

• 17:50Speaker 01

  So I got involved about two and a half years ago at Luther AI as part of the community, where we were like, let's build an open source version of GPT-3. Cause they, OpenAI stopped releasing stuff after their investment from Microsoft in 2019, because they said it was too dangerous, which is ironic, given the original founding statement of OpenAI. But again, it's their prerogative, right? Because again, technology that's powerful can be considered dangerous. And you have to sort of do that. So it was language models first, but then January of last year, when Clip came out, I actually built a system, that's when I was actually getting over COVID. I built a system for my daughter to generate art based on that. And it was amazing. So she created a vision board of what she wants to make. And then she made a description and generated 16 images very slowly, which were like a bit

• 18:35Speaker 01

  mushy and stylistic. And then she told how each one of them was different and the system interpreted that to generate 16 more, 16 more, 16 more. Eight hours later, she generated an image that she then sold as an NFT for India Code Relief. She raised three and a half thousand dollars. And I was like, this is amazing, especially because I have aphantasia. So I can't view anything in my head. It's a condition where you can't visualize anything. And I was like, something, I can visualize stuff. Wouldn't it be great if anyone could visualize stuff? Because the way that I've thought about things is that you and I doing what we're doing right now, talking, is the easiest thing in the world for humans to do, relatively speaking. Sometimes you need a drink, right? But it's still relatively easy. Written is harder. That's why we pay people to be writers. And image is the hardest.

• 19:19Speaker 01

  Creating art or PowerPoint is just really difficult and painful. But this technology can make it easy. So let's fund that. So last year I funded the whole space and all the notebooks and models and developers, like I hired them, I funded them. Gave them benefits, whatever they wanted, started building the compute resources. And there were a whole bunch of different models. Um, the stable diffusion model came about from latent diffusion, which came out of CompViz. Um, so that was a paper led written by Robin Romback, who's our lead generative AI researcher, and Andreas Blattman, who's joining us shortly. Um, that was kind of a bit of a breakthrough in high speed because they didn't have access to many GPUs, so they really optimized for high speed diffusion.

• 20:01Speaker 01

  Um, most of the advances in the sector, I think, can be credited probably to Catherine Crowson, Rivers Have Wings is her Twitter handle, who's our other lead generative AI researcher. Um, and again, she was just in the community and just was delighted to support her in kind of building these models, as well as other teams like the RU Dali team and others. Um, but then in about February of this year, kind of Robin messaged me and is like, we need to scale this up. I think it could be a big breakthrough. Agreed to it. And then the original stable diffusion released in August was under LMU CompViz. Um, so CompViz is the lab led by Bjorn Omen and Robin and then Patrick, who was at Runway ML, who's Runway ML is their lead generative AI researcher, were the two leads on that. So Robin is a stability and then Patrick there, um, creating it because

• 20:46Speaker 01

  the approach that I've always taken at stability, because we support communities doing all the models, is a collaborative one whereby our core team, infra team, academic, independents and others all coming together can build much better technology, you know? Um, and that's what happened with stable diffusion. A whole bunch of people got together, but it was really Robin and Patrick leading it. Um, and they pushed the boundaries and achieved amazing things. They took a hundred thousand gigabytes of images and compressed it down to a 1.6 gigabyte file that could create just about anything. And that was insane. And that was released August 23rd. And yeah, since then.

• 21:23Logan Bartlett?

  So what's been the growth? So the company was established. So, so, so just so folks are following, stable diffusion is the open source project of which you helped fund and and part of your team also, uh, started. And then, uh, the company around it is Stability AI. So when was Stability actually incorporated?

• 21:44Speaker 01

  About two years ago, when we were leading one of the UN AI initiatives. So we designed and architected that. And then it kicked off probably about fourteen months ago, saying, let's do all the types of AI. So right now we do all the types of AI from language models to protein to image and others. But Stable Diffusion is the most popular open source software in the world ever. So since launching August twenty third. It's received forty six thousand GitHub stars between our version one, which was this collaborative thing and version two, which was our highly optimized version that we ourselves released. Plus a bunch of tools around that. To give you an example, it's overtaken Bitcoin and Ethereum, which took about ten years to get to that level of developer interest.

• 22:26Speaker 01

  And when you add up all the stars of the ecosystem, it's now the most popular open source software in the world ever, just in three months.

• 22:34Speaker 01

  So the other models are amazing, like the language models from Alitha, which is one of the communities that we support and we hope to spin off into foundation soon. They've been downloaded twenty five million times, the most popular language models in the world. But this thing is just the most disruptive thing ever, and next year's it's going to get even more disruptive. It's what powers things like Lensa, which is the number one app on the App Store. I think they're making five million dollars a day. It's quite nice. And a whole bunch of other things.

• 23:01Logan Bartlett?

  Maybe tell people what Lensa is. I played around with it, but.

• 23:04Speaker 01

  Yeah, so Lensa or Dawn AI, you upload ten pictures of your face and then it puts you in all sorts of different like artistic variants and things like that.

• 23:14Logan Bartlett?

  We'll upload here my version of it for people to people to see, but it is it's super cool to see the the power of these things.

• 23:21Speaker 01

  But these things are getting, again, exponentially more powerful. So when we released Stable Diffusion August the twenty third, it was five point six seconds for an image on the highest end graphics card. Now it's naught point nine seconds for an image. In January, it'll be thirty images a second. It's a hundred times speed increase that we've managed to achieve, working with various teams around the world, which is insane for this tiny one gigabyte file. So what you just saw with Lenza, imagine if you could do that whole process 20 times faster.

• 23:52Logan Bartlett?

  I mean, it's super cool. Hopefully, people admire the picture of me that we showed on screen for YouTube people. But what are the use cases today that have people so excited in a practical sense? Like, obviously, it's cool to be able to do this in real time myself, but...

• 24:11Speaker 01

  Look, it disrupts the entire creative industry. In a year or two, we'll be generating whole movies in real time.

• 24:16Logan Bartlett?

  And what does that actually mean?

• 24:17Speaker 01

  It means you described that I want to generate a movie about, I don't know, Emad and Logan having a coffee at Starbucks or whatever. You input a few assets of our faces and then a short while later, you have a movie about them having a chat. And the chat is instantly generated as well about any topic that you want. If you give a practical example right now, there's a film that is shooting, I can't reveal details, with some very famous people. They had to do a photo binder with like 30 different actresses inside it. And those actresses were victims of a serial killer. It would have cost half a million dollars when you look at SAG daily rates, makeup, shooting, everything like that. Did it in two hours using this technology, saved the production half a million dollars.

• 25:01Speaker 01

  We're seeing companies basically bring videos to market 75% quicker. Sorry, 20, well, three, four times quicker. Video games will be generated, the assets for that even quicker as well. It's about 25% of video game budgets. So the people that are using this technology are just massively slashing creation costs. So there's a real enterprise solution version of that. For the average individual listening to this, this is the technology that will mean that you'll never have to build a PowerPoint slide again in a couple of years. Because you just describe it and then you'll say, make it happier or sadder or whatever. When you combine this with a language model and code model, and you'll never have to see any of that abstraction. You know, so this is crazy impactful technology.

• 25:43Speaker 01

  The fact that it goes real time, the, it's not just the image creation, right? It's the image editing. So we released an in-painting model, a depth to image that takes your face and puts it into 3D. And then understands all the lighting, so you can adjust your lighting dynamically.

• 25:58Speaker 01

  Upscaler, so you can go to 4K to 1K, just almost real time. Making that real time means that, you know, you could say, I want Emma to have a hat and then I want him to have a less bushy mustache. You know, what happens if his eyes are green? And it will do that instantly. It removes all the barriers to creation. I think people aren't ready for that. I'm not ready for that. And it's there right now. This is also the case when you see it go consumer, like I said, lens, it's gone. Our kids tell us about it and stuff like that. And there's other technology that's happening exactly at the same time. That will be just as disruptive like chat GPT, etc.

• 26:33Logan Bartlett?

  Yeah, I want to get into the chat side of it, but but while we're talking about image and video and all that. I guess I'll give you an opportunity to wax poetic and maybe a little philosophically about the implications of of what this actually means and what creativity. Because I assume anyone listening to this, like some of the things of the ability to make a movie in real time about the two of us having coffee and change your mustache and all that. It's a great tangible example, but maybe hard to understand why that matters. And obviously, there's the more tangible cases of the ability to green screen out backgrounds or the serial killer example in the movie that you

• 27:15Logan Bartlett?

  movie example you gave, but can you talk a little bit about like the power of of creativity and imagery? And what you think this this unlocks for people?

• 27:26Speaker 01

  Yeah, look, my mom sends me memes every day now using this technology about why I don't call her. And my guilt levels have gone up massively. You know, it allows people just to create anything. And the value of creativity versus consumption can't be underestimated. Like, one of the most effective therapies for mental health is art therapy for a reason. Because this is communication. And how many people listening to this right now believe they can create? Probably very few, but the reality is that everyone can. But they don't have the tools to, and they have barriers to it. Those barriers will be removed as of next year. You'll be able to create anything you can imagine, first in 2D, then in audio, then in 3D, then in video. And then you'll be wanting to share stories. Because I don't think it'll be like, you remember that scene in Wally where he's got like that VR headset? And he's all fat and stuff and everyone's in their own world?

• 28:11Speaker 01

  I don't think that's the case. People like sharing. We are story driven narrative creatures. And this allows us to tell more stories. And I think it gives people agency. Because, you know, again, like I've done lots of art therapy with people, like it really improves their lives. It improves your life when you are creating, no matter what it is. And then, again, too few of us believe that we can. We lose that childhood joy, right? Like when you're a kid, of course you can create. Then you get to your teenage years, you're like, oh, those people are better than me. I don't have time to do that. And then it moves away from that. And when you get to sad old folk like us, it's like, I have to learn to draw or paint or something like that. Maybe you do that on a holiday and then it's really rewarding. But too many people don't have access to that. I think, again, we've made that happen. So I think creation beats consumption.

• 28:56Speaker 01

  And now everyone can create. And so I think the world will be happier. Some people are going to use it in a douchebag way, but that's why we live in a society that has reticence against this. And this technology is pretty inevitable and it's going now. And again, exponentials are a hell of a thing. So we've got to get used to it where it impacts our industries and we're going to take advantage of it where it can make our lives better.

• 29:19Logan Bartlett?

  The inevitability of a lot of things you're talking about is increased productivity, increased efficiency, increased creativity, right? The flip side of any big gain in productivity is a loss of potential jobs or existing skillsets or responsibilities that people have. In your case of, if you speed something up seventy-five percent, there's some vector of time and so we're giving back time to people. But there's going to be an inevitability of, some people are going to probably lose their jobs over these types of things. How do you think about that in productivity gains versus, you know, some of the things that could create job loss?

• 30:00Speaker 01

  Yeah, so this has been the nature of technology, right? Technology has always led to productivity gains. And yet here we are at full employment today. Like I believe the number of photographers has increased by twenty-five percent over the last three years. iPhones are really good at photographs now, but yeah, there's still a job there. This is the thing I said about AlphaGo and Lee Sedol. The average level of Go player has gone up exponentially over the last few years. So I think this is augmenting technology as opposed to replacing technology. But there are certain areas where you have to consider what is the future of the rendering industry and other things like that, when you can automatically generate any type of asset almost real time. You know, um, so I think it will create and it will destroy. And this is the nature of technology. Again, the technology is not neutral. Technology is kind of this unmasking thing and it'll never be completely positive.

• 30:46Speaker 01

  You know, um, so yeah, that's why I think the important thing is given the pace of this adoption of this technology versus any technology that I saw when I was a hedge fund manager or I was a VC for a bit. People just got to get used to it and they got to really understand it because it's something quite alien and massively impactful.

• 31:04Logan Bartlett?

  I could make the, or I think people have made the case that um, this is so powerful that there's negative ramifications. Uh, the likes of which we don't even know what the implications of this are from a societal standpoint. And I think we've seen some people call for all these things to be uh, pulled back until we have a better understanding and put, can put in the right constraints and controls. I get the feeling you have a much more optimistic uh, take about human nature and also how these things sort out over, over time. Um, can you just talk a little bit about like your perspective on the, the um, the trade off between openness and accessibility with

• 31:47Logan Bartlett?

  uh, negative unintended consequences in general?

• 31:51Speaker 01

  Yeah, well, because the powerful technologies can do anything because these models are general few shot learners. As in they can learn a lit and then they can just do just about anything. It's a very powerful one, right? But like, I was thinking when people were saying that, I was like, so why do they want Indians or Africans to have this technology? Because actually it's an inherently colonialist kind of racist way to look at the world. Because when you ask them, it's when they educate them enough or they don't know better. Because there is this thing where there's this zeitgeist that tech people know better, but then it's like, nobody elected them. So they're self-appointed. So what's the answer to this? The answer for me was to move towards open, but that widens the discussion. So how many developers now are developing on stable diffusion?

• 32:33Speaker 01

  Millions. They're all voices now and they're people who weren't developing before. How many governments are talking about it? Just about all of them. All of the media studios are talking about it now, and it's in the public sphere. And there will be policy debates on this. And that's a good thing. You know, again, we have mechanisms in society to decide about these technologies and other things. And I think the overwhelming output of stable diffusion has been good. Like 4chan and places like that have had this for months now. Nothing bad has really come of it. You know, we've had technologies to do deepfakes and other things, but people haven't realised that it's quite so easy to do. Now they do realise. So my thing has been about, I think it's unethical to control access to powerful technology. This kind of echoes as well the cryptography debates that we had decades ago.

• 33:17Speaker 01

  Whereas like bad guys can use this to do bad things. What would happen if you'd had mathematics outlawed like back then? We wouldn't have cryptography saving us from the bad guys. You know? So I think there's a lot of red teaming because it is dangerous and because only big corporations can do this. And they were too afraid to release it. I think there was enough green teaming of what could be the positives from this, because like I said, how many people in the world can create next year who couldn't create because of us? Hundreds of millions. And that is a net benefit and happiness to society. But that isn't reflected in anyone's bottom lines, right? Well, maybe ours. But you know what I mean? So I think this is a very complicated thing, and people talk about it in three terms. There's ethics, and ethics is my personal and your personal.

• 34:01Speaker 01

  I don't think anyone has a right to call anyone else unethical unless they really understand that person. Morals are what we decide as a society. But my morals in the UK are different to your morals in the US are different to Indian and Chinese morals. And then finally, there's legal. And legal are these codifications of moral boundaries that we kind of put in. And we need to catch up on all of these. There's one final point I'd like to make is the alternative is that this technology is the pre-serve of large companies. You're mostly focused on ads. And it's really persuasive, this technology. So like, we've got human realistic emotional voices and faces and other things. What would happen if your chat assistant started whispering at you to buy something? What's the regulation legislation around that?

• 34:44Speaker 01

  What's the legislation around really large language models as opposed to our tiny models? That can get to human level? And should only big tech be allowed to use that? I think it should all be regulated. I don't know what that regulation should be. And I will give my voice to it, and I hope more people give their voices to it. Instead of it being just decided behind closed doors. So it's quite a complex thing. And I don't think there's any governance structure currently that I've seen that can handle it. I think that we should work together to put these things in place.

• 35:13Logan Bartlett?

  Now, the state of the industry today, there's obviously the large companies that have have some advertising related business, Google, Facebook, et cetera, Microsoft. And then there's the private companies kind of going after this opportunity. In my mind, there's businesses like we talked about runway or mid journey or some of those. But that the two fundamental companies, it seems, going after this, or at least when I talk to people I hear about, are you all and then OpenAI, who we've referenced before. How do you contrast your style and approach to what they're doing?

• 35:54Speaker 01

  So, yeah, I think OpenAI and us are the only independent multimodal companies. Multimodal being that we do all of the types of models, right? Again, runway do fantastic work on video, mid journey. David is super focused on kind of images and video games in the future that are streamed. That'll be super cool. We're kind of foundation layer companies, as it were. So we're building the building blocks that make this accessible. OpenAI kind of emerged, Elon Musk and a whole bunch of others wanting to build an open nonprofit for getting to AGI, this AI that can do anything to augment human potential. Had its ups and downs and changed over the years, but it's doing amazing work right now. But their objective is that generalized intelligence. And their model has moved more closed now because they think it's dangerous.

• 36:39Speaker 01

  Although they release amazing open source stuff, so they just released a good tokenizer. There is Whisper, which is one of the best things to turn this podcast into text, et cetera. So, yeah, they're selective about that, and that's their prerogative. Their model is data to the models. So you can fine tune the GPT, which is the language model, and soon I'm sure their image model. Hopefully they'll open source GPT three as well, sorry, DALL-E three, which is their new image model. Now that we've shown that it's rather to be safe, because they want to make their models better. And then they have a deal with Microsoft where Microsoft commercialized their models.

• 37:09Speaker 01

  And then funds them, which is a great, great, great kind of partnership. Our model is a bit different in that our model is models to the data. Whereby we're creating open source models that you can take onto your own code base or your own asset images. And we've teamed up with like AWS SageMaker for this. And then you can customize it to your own experiences. We don't really care about taking customer data and using it to improve our own models. Instead, we're like, scale and service is the way. You know, so if you want a customized version of the model, the best people to come to is us. And it'll be a million dollars a pop to train those things up. You know, actually more in some cases. If you want to scale it, scaling these models is hard, but we can scale your customized models for you. And again, we will have a fair deal on that one. So even though we overlap in certain areas, I think we have very different philosophies.

• 37:53Speaker 01

  Cause also our philosophy is getting this AI out to everyone and having everyone have their own personalized models. Versus building an AI that can do anything. And I think it's quite complimentary as well, because you'll always have a Windows Linux, you know, uh, kind of thing going on there. Oracle, MySQL, et cetera. So yeah, so I think that hopefully kind of finds it. I think the final thing as well is that their focus has been on language models, such as the amazing ChatGPT they've just released. With some image, whereas our focus is on media models. With some elements of language and others. I think in the respective spaces, there's a lot of people who are quite amazing doing language models. I think we're the only ones doing media models at scale.

• 38:35Logan Bartlett?

  In inherent in all of that, the approach and the philosophy is, is the open source nature of what you guys are doing, right? And as you think about the, the business model around the open source, you touched on the, the service and, and management that you can have with different customers. But how, how do you think about what the ultimate business model for stability? AI is going to be over time. Does it, does it stay around that? Do you sell different applications on top of the existing primitives? What do you think the, the end state of this is for you?

• 39:08Speaker 01

  We're fully vertically integrated, so we have our own product suite in Dream Studio Pro that we're releasing in January. Where you can generate entire movies and storyboarding and 3D cameras and audio integration with our audio models and things like that. Cause the models need to be used. We've also got integrations into Photoshop. And all the other kind of interfaces as well, where you can use our services and custom models soon. So I think that's a really nice place to be. It's the layer one for AI. And you know, we support the whole sector. So when certain API companies who aren't us got stuck on GPUs, we unstuck it. When mid journey was going, we gave them a small grant to get going to do the beta. Because we thought it was amazing to have this technology out there. So we really view ourselves as that infrastructure layer, picks and shovels, as it were, that other people build on top of what we do.

• 39:53Speaker 01

  You come to us if you want to have the vertically integrated best people in the world working with you. And every media company in the world and video game company needs that. And there's no other alternative. Because we build the models, we put them out there, and we also make them usable through our software. But you know, we're not going to have huge number of customers. It's going to be very selective, similar to a Palantir type thing. Um, and then the tail is where we collaborate with partners like AWS. We make our models available for everyone and we'll have more and more services around those across modalities.

• 40:26Logan Bartlett?

  How do you think about the uh, the difficulty of of model and technology between image and text? So obviously, there's, there's differences between Dolly two and stable diffusion. Uh, but I I know you guys are working on a bunch of different models and is the difficulty of of making language work harder than image? Uh, as as you think about the problem set, are they roughly the same?

• 40:51Speaker 01

  They're roughly the same. Um, language is a lot more semantically dense, which is why the language models are a lot bigger. So it's incredibly difficult to get them to work on the edge. With stable diffusion, we announced distilled stable diffusion, which is a twenty times speed up. Which will mean that we've sped up a hundred times since launch in August. So that means it will work in one second on iPhone without internet, maybe two seconds. That's insane. Language models cannot do that. But language models you can make great. So like I said, chat GPT is an example that's hitting all of the press around that right now. Um, so I think this challenge of getting it accessible is going to be the big one for us. And again, this is kind of our focus versus everyone else. Nobody else in the industry is focused on how do you get these things working on mobile.

• 41:34Speaker 01

  You know, because that's not their prerogative. Or how do you build an Indian version of this, et cetera?

• 41:39Logan Bartlett?

  And so do you think that specialized models will exist and thrive? Or do you think eventually these converge on large multimodal models? I guess, in other words, like, will there be significant leading models for just text and language and just images and proteins? Or or will it be one large model that's most effective?

• 42:03Speaker 01

  I think it'll be a mixture. I think every, if you kind of look at it, what happened is that the models got bigger and bigger and bigger. Trillions of parts.

• 42:10Logan Bartlett?

  Right.

• 42:11Speaker 01

  Unwieldy and inaccessible for anyone.

• 42:14Speaker 01

  Then it turned out that actually we weren't paying enough attention to the data. So DeepMind released something called Chinchilla, which was a version of GPT-3, which is 175 billion parameters. But in 67 billion parameters, the app performed it because they just trained it longer. But if you look at the actual import of that paper, it was just you need better data and training longer. So we don't really know how to optimize these models because they were so big and so compute intensive. Because they cost millions of dollars each that we didn't really see what the differentials are from data, from training and others. There's a lot of model optimization to go. One of the big breakthroughs now, though, is that we've moved from just deep learning. So these big supercomputers squishing the data and move it back and forth across all these chips to now introducing reinforcement learning with human feedback.

• 42:57Speaker 01

  That's where you see how these models with all the little neurons, which are these principles they've learned, light up when humans actually interact with them. And you use that to create more specific optimized models. So OpenAI, again, like leaders in this field, they created InstructGPT by figuring that out from GPT-3. It went from 175 billion parameters to 1.3 billion parameters with just as much performance. And this is one of the things that I think will really drive it. The combination of deep learning and understanding how humans interact with these models to make better models and to get better data to build better models again. You know, we're reaching that point now of rapid iteration and feedback, and that's what we saw with stable diffusion. Got 100 times speed up.

• 43:39Speaker 01

  You know, so it's a hundred times faster than Dali two as of like now, basically, we'll release it shortly. Um, but there's more to go. And so I think, again, none of this really makes sense, to be honest, Logan, like the fact that a 1.6 gigabyte file can contain two billion different concepts and create just about anything now, pretty much photorealistic, doesn't make sense. It's insane, but it's there. And the fact it can run on iPhone without internet doesn't make sense. You know, it's orders of magnitude better. And the question is, how are we going to react to this when it happens for every modality? Because some modalities are more difficult than others, like 3D is very difficult. But maybe they'll figure out how to make it easy. I don't know. You know, video is difficult. We're figuring out how to make it easier. I would, I would note that 80%

• 44:24Speaker 01

  of all AI research is in this area. And I reckon $100 billion of investment will go into this area to accelerate it even more.

• 44:32Logan Bartlett?

  And as you think about your, your core use case and customers, you mentioned closer to the Palantir model, do you think it's mostly going to be focused on, on media and content are going to be kind of the core businesses that you service? Or are there other use cases or target customers that you're excited about?

• 44:51Speaker 01

  Oh, yeah. I mean, like, I think we can disrupt the whole of pharmaceutical industry and healthcare. Um, and again, it's an area that we know very well, and we've got protein folding and other things, but these models can be applied to save billions of dollars there.

• 45:03Logan Bartlett?

  And why is that the case?

• 45:05Speaker 01

  Well, because the whole area is kind of classical systems are a Gerdic. So they treat everyone like distributed normal, like that, right? It's like a thousand tosses of a coin is the same as one coin tossed a thousand times, but people are very individualized. We didn't have the systems to be able to have personalization and understanding of principles. And we've got them both at the same time now with this technology. When you apply this appropriately to healthcare and bio, you have massive breakthroughs and there are drug development companies and others doing that. But they're all building their own infrastructure when they should be using a unified infrastructure. And this is where you open source is super powerful. Because I'll give you an example, by releasing our model and having the traction, Apple optimized it for the neural engine on the

• 45:47Speaker 01

  M1 and the other architectures. It's the first model ever to basically have that. You can set the standards around this, then you can really drive forward the sectors. And from a business perspective, again, we are the business that does this. There is nobody else. That's multimodal. There is nobody else that is media focused initially. So media is our number one thing, media and video games. And so if anyone wants to have the best, they come to us. But we can only work with a few entities, but that's okay, because we can have entire massive business based on that. Just like Google has a great business based on ads. You know, these other sectors, we can do the sectors one by one by one. But I don't see any sector that's not affected by this technology.

• 46:27Speaker 01

  The only question is for us as a business, who do we work with partners to go out to with this infrastructure? And who do we do ourselves? I mean, it's kind of like what the promise of Web three was many years ago. But Web three was always an economic incentive that was trying to be bootstrapped to real life use cases. Whereas this is real life use cases right now. They're delivering value right now. I think that's why it's exploded and next year we'll go even bigger.

• 46:51Logan Bartlett?

  One of the interesting elements of this is, is it's really coming. I mean, historically, automation and AI has, has been assumed to serve at the, the simpler levels, right? And move their way up or the more manual levels and move, move their way up the stack. This is actually fundamentally shifting that paradigm and coming at it from the, the creative, uh, more, uh, more knowledge worker based, uh, systems and processes and actually, uh, potentially automating away a, a bunch of, a bunch of those different jobs. When you think about like, no industry is not going to be impacted by this in, in some way. I'm just fascinated to hear you riff on some of the different use cases or industries.

• 47:36Logan Bartlett?

  Have you seen it applied, uh, in, in any ways that were unexpected to you? Since this has been out in the world in August, that, that would be tangible for people to hear about.

• 47:47Speaker 01

  I mean, it's been applied in crazy ways. People have used it to create 3D VR simulations instantly. It was used to create synthetic data on lung scans to identify cancer by Stanford AMI. Because they didn't have enough data sets and they created more data sets like the ones they did. Yesterday, there was something called Riff Fusion, uh, whereby they took spectrograms of music and trained it on that. And now from those spectrograms, you know, the little things, it can generate brand new music of any type. Which is a bit crazy. So I think that nobody really knows what the long term implications of this are. A lot of people think that it should just be words and images out or words

• 48:30Speaker 01

  in text out. But the real impact's going to come when people really sit down and think about which parts of the creative process, the constructive process, my office process, with a little bit of some sort of entity that understands the nature between structured and unstructured data and the barriers between that and blurring that be super useful. And I can't really think of many things that aren't disrupted by that in the knowledge workspace. In the manual workspace, it's more difficult because you needed to have robotics and things like that, high capex, right? There was no capex required to do this at a base level. When you want to build your own custom models, yeah, it costs millions of dollars. But only a few companies will do that, which is why our focus is on a few companies. And for everyone else, it's just making these models usable.

• 49:15Logan Bartlett?

  What concerns you the most about about all of this being out in the world now from a societal impact standpoint?

• 49:22Speaker 01

  I mean, look, it's the unknown unknowns, right? Nobody really knows what's going to happen. The bad guys already have this technology, you know, for a number of reasons, right? And I don't know if we'll be able to catch up enough with the society to the bad actors who have better versions of this technology. But then also, some of these knock on effects, like, you know, anyone can create anything, anyone can write anything

• 49:47Speaker 01

  instantly. Like, there's no more barriers to this. What are the knock on knock on effects of this? I don't know. Nobody knows. So this is something that worries me a lot. But then the other part of me is like, the alternative is, this technology is only controlled by large organizations, and they're full of good people, but they do bad things. And they will use it to service more ads. Or we could use it to activate humanity's potential by bringing this to the world and having an open debate. And so when challenges come up, we can react to it together. You know, so that's how I've tried to mitigate against that. That's kind of my approach to philosophy.

• 50:20Logan Bartlett?

  Who's the most forward-thinking in terms of the the risks and trade-offs around all of this? Because as you mentioned earlier, ethics, morality, laws, all of those things are are very, very different. And what UK versus US, I think I've heard you reference, like, there's no absolute moral framework for things in the world. And like most people that do harm or ill, as we view it, they can talk themselves into anything and believing anything. Right? So I guess I'm interested in in who's at the forefront of thinking through some of these some of these things? And and what's going to be the governing body in your mind that that actually comes to these types of decisions?

• 51:06Speaker 01

  I've been very disappointed. I've not met anyone who's really thought this through. The classical thing is either massive techno optimism, everything will be absolutely fine. Or massive ultra orthodoxy, this technology is too dangerous to ever release. There are very few people who are kind of stringing the line between, I think, UK government's probably most forward-thinking on this. The European government is the most regressive. They're looking to ban general-purpose artificial intelligence and be the regulatory leaders, which is stupid. Europeans will fall behind. The US is trying to figure out where its stance is on this in terms of governments. Like I said, even on the individual side, I just think that it's really complex. And I can't think of a government structure that can handle this because one of the questions you like, do we give this to the Linux Foundation?

• 51:50Speaker 01

  Or something like that made decisions? Not really. These are de novo things and you have to kind of make decisions around this.

• 51:58Speaker 01

  But I think, you know, a lot of the press against us, we had a lot of positive press, a lot of negative press. But at least it's out there and I like that it's out there because it means that people are having really strong discussions. I think we need to have more structured forums to have these discussions in a proper way, not in an emotional way. And I think that will happen, hopefully, as these things get more exponential and we will host some of them. And we'll invite as many people as we can into this. This is also why, like I said, our communities around language, around bio mail and others, we're spinning into independent foundations to handle small parts of this.

• 52:30Speaker 01

  So that and we're inviting everyone in on that. So we shouldn't be making these decisions and I should anyone else about what a benchmark model is. But then for the overall guiding thing. Yeah, nobody really knows, unfortunately.

• 52:43Logan Bartlett?

  The last four months of your life, or five months of your life, or whatever it's been, I'm sure have been pretty unusual. What's it been like at a personal level? One, to have the largest open source community in history, or at least trending in that way? As well as, I mean, now you've become something of more of a public figure than you were in the past. What has this been at a personal level, managing those two things?

• 53:10Speaker 01

  It's been really tiring, stressful. I mean, look, I never wanted to be a public figure. I got Asperger's and ADHD, I hate public. But like this needed to have a figurehead and someone to lay the blame on. And I also got the positives from that, you know? Built a great company, had lots of failures in the past. I'm just trying to do my best because unfortunately, a lot of this stuff keeps centralizing. So I keep trying to give away a lot of authority and it comes back to me. And that's really heavy and it's a heavy burden to bear. But like I said, there are positives that counterbalance that. I hope I do the right thing, but the amazing thing now is that some of the smartest and best people in the world in various sectors are reaching out to us and joining stability. So I think if we improve as an organization and build a great organization, people can be part of, you know, like Google in 2012,

• 53:55Speaker 01

  then maybe this can be dispersed amongst really intelligent people who've got good hearts, working in the right way. I do think we need to be more transparent as well, like there's a tendency to keep things closed for a variety of reasons. And so I hope we can become a really transparent, great organization, full of great people. Because then I can go and finish like my video games and things like that and then take a backseat.

• 54:17Logan Bartlett?

  Who do you turn to for, I'm sure you have tons of people willing to offer advice or opinions? Like what is your, the group of people you keep around the table that keep you level-headed and balanced? And help you drive towards the right north star? Or does it inevitably end up being just your gut at the end of the day? In terms of what good actually means in this world of no absolute moral framework around that stuff?

• 54:44Speaker 01

  I mean, that's a complex thing, right? Kind of getting into metaphysics and things like that. Now, I've got my friends, you know, put a board together because they can give me advice. Got some excellent people like Sri and Jim on there for the business side of things. And it's the team as well. Again, I need to communicate better with the team, but they tell me very directly when I'm being stupid or overall, or, you know, sometimes a C V C E O overhypes things because he gets excited. So there's plenty of checks and balances because also I'm quite an approachable person, I hope.

• 55:15Speaker 01

  But yeah, like I said, ultimately, founder-led companies can only sustain for so long. And we're in this transition point now where we're going to become a process-driven company. And we have to, not from a business perspective, but also from an important perspective, given the impact of what we're doing and our place in the ecosystem. Like, again, build a great company to help a billion people. But at the same time, this is an important point in humanity, given this technology, like I said, next year is going to take off everywhere. Like, every single graphic designer and every single person will use stable diffusion in somewhere or another. Every single person doing their homework will use Chat G P T. You know, these are big changes that are coming through.

• 55:53Speaker 01

  And it's been building, I think you've seen it building. But now it's this breakthrough moment that's occurring.

• 55:58Logan Bartlett?

  Why do you pick, why do you pick next year as the time that that this is all going to happen? Is it just extrapolating the exponential growth curves?

• 56:06Speaker 01

  Real-time, stable diffusion, apps like Lens are hitting number one on the App Store, showing the value of creativity, like 500 million run rate. And like I said, the final element that was Chat G P T, whereby every single smart kid I know now is using it to do their homework, at least 80%, right? Like, it's good enough, fast enough, cheap enough. And that is the takeoff point.

• 56:27Logan Bartlett?

  Great. Well, Ahmad, thanks for coming on and doing this. This is super fun. It gives us a lot to think about. So, so thanks for coming in and answering, answering all those questions.

• 56:37Speaker 01

  My pleasure. Thanks for having me. Cheers. You take care.

• 56:44Logan Bartlett?

  So that'll do it for the 46th episode of Cartoon Avatars. Thank you to Ahmad Mustak for coming on and having that conversation. And thank you to Andrew, Justin, Jenny, Rashad, and everyone that helped out on this episode. Look forward to hearing, to seeing everyone next week on the 47th episode. We have a guest that I've long admired and someone that I was definitely excited to have on, so you'll, you'll definitely enjoy that one as well. Thanks everyone for listening.