Outcomes of an AI Future W/ Emad Mostaque | EP #55

24:41That's why you have to hit the blog button because it's a little echo chamber, and if you dozen people can have that impact upon you. I mean, it's going to be very interesting to see the way that our adult minds and our kids' minds evolve over the next five to 10 years with the emergence of this new, more powerful, personalizable technology. Yes. It's either controlled or controlled. Everybody, it's Peter. I want to take a break from our episode to talk about a health product that I love. It was a few years ago. I went looking for the best nutritional green drink on the market. I went to Whole Foods and I started looking around.

25:18I found three shells filled with options. I looked at the labels and they really didn't wow me. So I picked the top three that I thought looked decent, brought them home, tried them and they sucked. First of all, they tasted awful, and then second, the nutritional facts actually weren't very impressive. It was then that I found a product called AG1 by Athletic Greens, and without any question, Athletic Greens is the best on the market by a huge margin. First of all, it actually tastes amazing. And second, if you look at the ingredients, 75 high quality ingredients that deliver nutrient dense antioxidants, multivitamins, pre and probiotics, immune support, pathogens, I personally utilize AG1 literally every day.

26:04I travel with an individual packet, my backpack sometimes in my back pocket. And I count on it for, you know, gut health, immunity, energy, digestion, neural support, and really healthy aging. So if you want to take ownership of your health. Today is a good day to start. Athletic greens is giving you a free one -year supply of vitamin D and five of these travel packs with your first purchase. So go to AthleticGreens .com -slash -moonshots. That's AthleticGreens .com -slash -moonshots. Check it out, you'll thank me. Without question, this is the best green drink product, the most nutritious, the most flavorful I found.

26:43All right, let's go back to the episode. So we're sitting here in the middle of Hollywood. I can see the beautiful Hollywood hills, Hollywood signs around here someplace. I just over there, okay. And the news today is Green Actors Guild. Everyone's gone on strike. There is great fear, pain, concern. And you know, you called it when we spoke six months ago. So what's going on? How do you view it? And where's it going? So I think the advances in media or artificial intelligence have been huge. You've now got the Drake AI song on my favorite ice ice matrix, where the matrix characters sing ice ice baby.

27:26Yes. You've got real time rigging. You have real time special effects. You've got high definition creation of anything. What does that mean? it means that the whole industry is about to be disrupted because the cost of production reduces. It was reducing in some ways anyway and we've seen this move where the cost of consumption went to zero with Napster and Spotify and then the cost of creation started going to zero with Snapchat and TikTok. And now we have a question of what are the defaults going to be now? and consumers only have one limited currency in that's their attention. I think so, but consumers are willing to pay for attention.

28:09They want to pay for quality. Well, either premium media or current or you know, otherwise. Because video games, for example, started as a $70 billion industry 10 years ago. The average score went from 69 % to 74 % on Metacritic over that period and now there are 180 billion dollars. Movies? I pay for some of that with my 12 -year -olds. Movies went from 40 billion to 50 billion but the average IMDB movie rating over the last 10 years has been 6 .4. It has not changed. So you're not producing more and so they're not consuming more. So there's a question. Will this technology enable an increasing quality because it raises the bar for everyone?

28:51Yes. There are more movie makers, so they're more excellent movie makers and the best movie makers become even more excellent And we've democratized the tools on from an iPhone to you know the tools on your on your Mac And of course YouTube has disrupt was the first major disruption of the it was something like mr. Beast has a much bigger audience and CNN now Yeah, but that comes down to what is the key point here distribution? You can make it stories but if no one hears them And that's what Disney has as its benefited distribution to its theme parks, to its products, to its channels. It creates a shelling point, but then if we will have our individualized AI, it can find us what we need and sift through the crap.

29:38Maybe Hall of Distribution flips on its head in 5 years as well. So I can't imagine, you know, I keep thinking that the future of movie consumption is me speaking to my version of Jarvis and saying, you know, I'd like a comedy. I'd like it starring me and three friends of mine and somebody else and for 90 minutes and set it in, you know, some setting and go and have it auto generate a compelling story that is either because I'm involved in it, I'm in wrap -shirt in it or because my favorite stars are all together in unlikely place. How far are we from that kind of future? That's probably three to five years.

30:25So in that case, it's not going to be cheap, but it'll be there, and then it'll get cheaper and cheaper. In that case, there is no distribution needed, but a person calls up whatever they want. I think there is distribution needed, but in a different way, music, we have a variety of different music sites, but how do musicians make their money now. It's not Spotify. A million views gets here a few thousand dollars. T -shirts, merchandising, global stories because there is that future of the Wally type of fat guy sitting with his VR headset. It's kind of depressing. Just like the Apple Vision Pro Advertisement, so I found kind of depressing when his kid's right there and just puts them on.

31:02Love it dystopian to me. I think it's more a case of there are certain stories that everyone wants to talk about. Like here in Hollywood, what do we have this week? We have Oppenheimer and Bobby. Oh my god. I'm looking for it at Oppenheimer. I will not watch Bobby. Sorry. Would you watch Bobbyheimer?

31:24Only... No, I won't go there. Well, I mean, we can take it. We can put the scripts into Claude and see what it comes up. That would be hilarious. Yeah, hilarious. Kind of Bobbyheimer. But again, like, people are talking about the Bobby movie because it's, you know, not, she comes to the real world and she has challenges. It'll be a hit. People will talk about Oppenheimer because again, it will be a hit. These are produced hits. These are produced hits. Just like I saw BTS with my daughter, I don't know BTS, they're Blackpink. Oh, God, she'll kill me. In Hyde Park a few weeks ago, the biggest K -pop band out of Korea, right, completely manufactured, but lots of fun.

32:00So on the one hand, what you have is what you describe your personalised things. That's kind of like McDonald's. What's the job to be done? The job is comfort. The job isn't to listen to someone else's story and expand from there. It isn't a produced Michelin Starmier or just a nice restaurant. One that you can talk about to others like you can cook ingredients yourself at home as well. Sure. But humans do like these bigger stories, but then the nature of funding of musicians changed. It was about merchandise, it was about kind of a lot of this other stuff around tours. I think the nature of movies like that.

32:41The business model has changed. I mean, that's the most interesting thing about exponential technologies is it's changing the business models and I keep on, you know, advising my entrepreneurs, it's reinvent your business models more than anything else. You have to always look across the landscape where is the value peaks and then you're sitting then you're intermediating something and you're offering service value. You know, like there was that great quote who is it from the CEO of Netscape. All values created by aggregation or disaggregation, bundling or unbundling. And you think about how the landscape going to change now as intelligence has moved to the edge.

33:23What was once in the hands of the studios and the high priests of media suddenly gets pushed to the edge, whereas value then it changes it, flips it. So let's go, let's go, what's going on right now? So the screen actor's guild is on strike because of why? Screen night, I mean, it's better wages in general, standard stuff, but now there's this rapidly emerging fear of artificial intelligence. So one of the proposals that came in from the other side was that basically all the extras, all the actors, they sign away their rights so they could be used in AI. So they get a day of wage to get scanned and then they can be used for the rest of that studio's life in producing background actors.

34:09And they were looking at this, like, oh my god, wait, you can do that. If you must be able to even realize you can do that. You know, the script writers are saying no AI generated scripts, which again is a bit weird when they're all using Grammily and things like that, which was AI. How are you ever going to tell where do you draw the line? Where do you draw the line? Because this is a technology that's coming so fast and is so good that it's almost not like technology at all. It's just very natural the way it emerges. And so you will get to some sort of agreement because the big act is kind of there.

34:43but the defaults that are set now reverberate. So I think that's an important point that you just made. The decisions, the policies that we create today are gonna take us down one path or another for the decades to come. Yeah, it will affect the whole of Hollywood what's decided in this moment here because there won't be another renegotiation for a while. And so again, how does an actor create value? A top actor has a following. But up in Cummers, how did they break through? What was the apprenticeship to him? What does a movie look like in five years? Even if you agree as Hollywood, not to have any AI.

35:26Let's just say you have this kind of do -installed but Larry and Jihad and say, no AI, you know? And someone will make a movie about no AI in Hollywood. What do you do when the Chinese film studios start releasing product, start raising products faster than anything. You can make fine in every language out there. Literally every language. We have technology now. Maybe this is part of what you, what is possible now? We can translate Peter's voice into just about any language with his voice. It's not a voice match. I match my lips and movements exactly. Match your lips and movements exactly. I'm sure the podcast will be in every language by next year.

36:02And again, it will be in our voices. We can make our voices sound more confident. We can take his mannerisms right now and transplant them onto my mannerisms so we match so we can reshoot scenes with style. Yeah, we can turn them into a robot in a few minutes and in fact maybe we'll do that in the form of the post. So we'll have a little scene of him becoming a robot. These are all technologies that are here now and it transforms fundamentally the nature of filmmaking because you only need one shot. But don't the filmmakers who are standing up for their rights to not have them basically demonetized and digitized, the other option is for Hollywood to just create complete artificial characters that they fully own.

36:46Yeah, and you've seen this already with some of the kind of V -loggers and others that have emerged have Asian, particularly fully AI -generated characters. And you can have entire mythos around them. And you can say, make it the most attractive Italian guy I've ever seen. That's Broody and this and that. How you gonna tell it from a human? I mean, these characters can be completely unique. And it's far more profitable for the studio to use that digital actor. So there is a disruption coming at every level. Because the research to revenue pipeline has become so tight and it sets off a race condition whereby you could already produce two movies a year suddenly you can produce 20 And then you can actually Like we do a B testing in subject headlines you could create 30 variants of the movie and see which one actually is the best Yeah, and I mean again you can say make it more or make that speech roar and more emotional, it will adjust the voice to make it roar and more emotional, right?

37:51And because you're using such a large data set, so you know, and we, so we made all our data sets open, and then we allowed opt out, whether you're only company in the world to allow opt out, because it was the right thing to do. So we had 169 million images opt out of our image data sets for music, because there's different copyright laws. We have one of the first commercially licensed music models coming out. So respect for that, But if I'm an artist and I go to the Louvre to be inspired and then go back and paint and I've been inspired by DaVinci and I start painting in a style like DaVinci

38:31Where's the difference there? Where's the difference and this is the reality even though we've done that By next year, probably by the end of the year, you will have models that have zero scrape data or human art. They will all be synthetic. And you'll be able to bring your art to it. So there's something Google just released called Star Drop. There's hyperdream, there's hyper networks. So hyperdream booth. You take one picture of yourself and the entire model trains to be able to put you into anything, even if you're not in the mall. It used to take minutes, hours. Now it's just one picture. You know, similarly you can bring any style and it will just imitate that style.

39:10And so all of a sudden the models themselves, it doesn't matter what they're trained on because there's no human endeavor in those models and then things like compensation for artists and others, as you said, become a bit mute because all of a sudden you have these amazing stories told by really convincing amazing actors who may or may not exist and how are you ever going to tell the difference? So what's your advice? Let's parse it here. On one side, what's your advice for Hollywood and for actors? And on the other side, I want to ask your advice for artists. Because this is about mindset. This is coming at us at extraordinary speed.

39:52There's no stopping it.

39:57There's no slowing it down. and so you've got to deal with reality. You're doing reality again, it's inevitable. Even if, again, Hollywood says no AI, the AI is coming from around the world. So what do you do? You think, oh, well, my audience suddenly became the whole world. Yeah, that's a big deal. You're like, what am I actually known for? Is my acting skills? Well, I will still get these things. You're an up and coming actor. You say I need to build community. I need to kind of shove something more than that. Because again, my acting skills in some areas can be transplanted, but what about real life shows, you know, what about these things?

40:34It does throw up the entire thing and adjust it, but then musicians have had to have that adjustment. They used to be able to make money on their LPs, and then all of a sudden they had the naps to Spotify moment. There is more protection in music as well, because you basically, according to the Robin Thick versus Marvin Gay Case, there is an element of style protection. in that, that doesn't exist in visual media. Yeah. And probably won't, because the other part of this is, if you're expecting governments to regulate, how can they, when there is a global competition going on? They will lose competitiveness to other countries, and you'll have regulatory arbitrage.

41:16Yes, and that is something across every industry that's going to be happening. Yeah, I think, you know, the concept of, you know, United Artists, It was originally a collective of all the artists. That makes a lot of sense now. I think you have to think about an element of collectivism to share the excess profits because what's going to happen is movies will get cheaper, profits will go up. You need to support each other as a community here and think again as a community what is our story for the next one three five ten years? Because all of this is going to happen quicker than it takes to make the new Avengers movie.

41:53And quicker than regulators are able to regulate. And again, the regulators almost certainly won't regulate because they will start falling behind their competitor countries. I mean, if we look at the internet itself, as it, you know, the media industry never expected the internet to have the disruptive impact it had. And had it known, it probably would have tried to get regulators to have slowed it down or blocked it. The speed is too much. But then also, again, I gave the example earlier, the video game industry has gone from 70 billion to 180 billion over the next. So it's lost to me. Can we increase quality?

42:28Interactivity. OK. Because games are media as well. The media industry is increased in size, the way that value has been redistributed. Value will be redistributed again now. And again, it's like, what is an AI -enhanced actor? If you're an actor, what's an AI -enhanced photographer, filming? Think about your jobs, the tasks that you do, And what can be augmented if you had a bunch of really talented youngsters working for you, right? You could do more you could be more But then it means the bar is just gonna keep on raising Let's turn to a different Industry that's going to change and we had this conversation in our last Podcast and on stage at about it's 360 which is coders Coding is changing dramatically What are your thoughts there?

43:20So, when I started as a programmer, I was writing enterprise level assembly code for voice over IP software. Wow. And to switch. It's some of the largest chunks of code out there. Yeah, well, it's very low level code. We didn't have GitHub. We just got subversion the following year. Programming these days is a large, like Lego, because what you have is kind of you have a very low level, but then you have levels of abstraction until you get to PyTorch and some of these other languages. So you have to compile lots of different libraries because you're making easier and easier. Human words are just the next level of abstraction there.

43:55But the nature of coding is going to change. And so the coders that are coding traditionally today around the world, how will they be using and working in this industry two to five years from now? Well, again, there will be no code that doesn't use AI as pasta workflow. I mean, I think that's the important thing. It's not like code is going to go away. They're going to be using a new set of tools. The expectations will rise. The amount of debugging, unit testing, all of these things will decrease. Because how much time did code is actually spent architecting? Very little upfront. It's more about understanding information flows, but architecting these things.

44:37It's about having feedback loops to understand customer requirements. data bricks is a $38 billion company, those data lakes. So it takes your data to organize it lies right, structured queries. You have to write queries. Now you just talk to it. And it just does it. Microsoft introduced the same thing. I mean, I can't wait for that in the field of medicine, which I want to talk about next. But you said something earlier where you can imagine there can be a billion coders in the future. Yeah, because all the barriers to creating programs disappear. Yeah. So it's not that there are no programs, there's no programs that we know it, because there's a billion programmers.

45:16Yeah. Everyone is a programmer, nobody is a programmer in a way, because it just becomes a matter of course. I want to make software that does something and reacts in these ways and looks like this and adapts like this. And then it comes to you and you're like, no, that's not quite right. I want this moved over. It happens almost live. this feedback loop. It's as we talked to JGPT for creating a paragraph that describes something we want, we modify it. I mean, like, Chant GPT -4 is a good example because to write an integration to something like Chant GPT -4, it used to take days, weeks, an API, an application protocol kind of interface.

45:55Now what you do is actually you tell it, the schema, you tell it kind of what you should do and it writes it automatically. literally within like a few minutes. And then in a few hours, you've integrated into it. I've been talking about a future where we all have Jarvis, Iron Man, I love Iron Man as a movie, it's one of my favorites. And Jarvis is basically your personal AI that is, it's a software shell. It interfaces between you and the rest of the world. You ask Jarvis to do something and it knows how the tool paths on a 3D printer, you can hop into a jet and interface Jarvis with the Jets computational system and it'll fly the Jets for you.

46:34I can't imagine that that is really far from now. Yeah, let's hope that doesn't fly planes just quite yet. Okay, we'll put the Jets aircraft aside. But the ability to, for it to become your best friend and confidant, know your needs and desires, shape the world to your comfort and being able to help you, it's the ultimate user interface. Well, I mean, this is why a lot of the chat bots, character AI and others have become so popular because it'll never judge you and it's Approach that human level now You know, and again, it is the ultimate interfaces may be chat, but as more chat is chatting context It's understanding you he'llistically no human could do that because you know even if you hire a whole team They're not going to be with you 24 -7.

47:22This will be with you 24 -7. I think the key thing here is empathy yes, because jumping head bit to medicine. Google had their Med Palm 2 model. The papers just come out in nature. A, it outperforms doctors on clinical diagnosis, which is crazy for a few hundred gigabytes of a file. B, it outperforms doctors on scores of empathy. I found that amazing. And totally logical, it doesn't judge you. It doesn't judge you, but then a doctor has split a million ways and they're tired and they're grumpy or this or that. Some of us get good doctors, most of us don't. Some of us get good teachers, most of them.

48:01I'm not saying education is bad because the teachers, so many teachers try so hard, but their attention is bit 20 ways and they're underpaid. You know, I'm not saying that programmers will be, the H programming will change because programmers are bad. There's so many hard working programmers. It's just again, the nature of these things will change when you can scale expertise. And everyone has expertise available to them on tap. I've been on the stage, you know, just pounding my fist saying, listen, it's going to become malpractice to diagnose someone without an AI in the loop within five years' time.

48:35And probably in some areas, it'll be inappropriate to not yet illegal to. And then at some point, soon after that, the best surgeons in the world are going to be humanoid robots that have every possible, You know, it's real, you know, variation, every possible, you know, history of surgery, and they can see an infrared and ultraviolet, and they haven't had a, you know, argument that morning with their husband or wife, and it becomes the best. And these are demonetizing and democratizing forces for health. There must be deflationary as well, but I know I agree completely with this because ultimately what's going to kick it off is is your doctor AI enhanced?

49:21Yeah. Lower insurance premium, lower copay. Yes. Because there'll be real economic incentives. Has this been cross -checked by the technology? Yeah. Reduced the cost. My favorite subject is you know you probably know this, how many medical articles are written in journals every day? I don't know. It's a 7 ,000. Wow. And it's like, how many of you're Dr. Red today? You know, and there may be that one breakthrough that happened this morning that is the key for your diagnostics. But I mean, even if they've read it, right, like absorbing it is one thing, having the mental models. These are constantly, this way in comprehensive authority to not stay, which is what this technology allows to happen.

50:03And you said things like, we've already seen some surgeries, can do you done better by robot surgeons and human surgeons, it'll be all surgeries. Yeah. It will be all surgeries soon enough. like the robotics advancements we've seen. This actually goes back to your point of, you know, the artist going to the Louvre and seeing the DaVinci and then taking inspiration from that. What are we gonna do with all of these optimists, robots from one KX, robots and others? Are they gonna have to shut their eyes when they see anything copyrighted? I'm curious. They're just gonna have accents everywhere. They're like running into each other.

50:33Everything's gonna be black lined out. Yeah. So medicine is changing dramatically. What other field are you seeing and seeing people need to wake up and see what's coming? So I mean medicine Education are kind of the two big ones. I think but can we move this to side right now again? We've seen programming the entire nature program will change media the entire nature of media will change from journalism to filmmaking But anything that basically You could do with someone from Asia on the other side on a computer screen will change Yeah. So education, because today education hasn't changed since the one room classroom, half the kids are lost, half the kids are bored.

51:24You're teaching to a series of tests. You're teaching for a industrial era world. And people learn differently. People have visual and auditory and tactile learning skills. and let's face it, we don't celebrate our teachers, we don't pay them well, we don't have the best of them coming into the classroom. And they're sad. I mean, this is a thing, like there are happy classrooms with happy teachers and other things, but learning should be a place of positive growth and joy. It should be fun to learn, and it should be fun to teach and fun to teach us. I think this is both of them, because what is the nature of a teacher in five, 10 years?

52:05Let's say ten years. I'm going to let's... Again, ten years is the crazy short period of time. I mean, I asked you this question last time, you know, how far out can you predict what's likely to come? What's your singularity boundary condition? I'm curious. What are you seeing as? How far? Three or five years. I mean, yeah. I got to Dubai at a month stage and can you talk to us about, you know, what the world's going to be like in 2050? My answer is no. I can barely talk about 2030. It's everything ever at all at once, lots of S -curves, acceleration. These are inevitable now. Now we've broken through these things and now open AI, now I've put 20 % of their stuff to alignment because they're basically saying that their view is five years out.

52:51Elon Musk just said six years out for the Elon's relationship with time is always a bit far. Just like self driving cars. Self driving cars are literally here now. Yes. You can get in one and it will drive you around San Francisco, London or Weimo will do that for you, not my Tesla. Not your Tesla. I know. I keep up pushing the button, but it doesn't. But I know that she is here. And I was like, oh wait. What? This is a thing. So again, 10 years is a dramatically short period of time for education, which has been the same for a century. Yes. But the thing is, it is inevitable that every child will have their own AI.

53:25Right. So your 12 year old will be 22. When they get to 22 and they come out of, let's say university, still around, university. If university is still a thing. Yeah. They will have their own AI that's learned for at least five years about them. Yes. That can fetch them any information in any format of any type and write anything. Yeah. Or create any video or movie for them to, that's a crazy thing. There were concerns that Wikipedia would remove wrote learning and things like that. Google would do the same. and maybe there are kind of again, like appendices that shivils. I don't know, no, it was the thing that shivils, the, you're appendix is right.

54:04But you know, like, I was at gallbladder thing. But this is the thing, like, you might have some vestigial parts of your brain. The entire human brain will be rewired. You must assume that every child will have their own AI, how that AI is driven is different, because any child that has the AI will dramatically outperform the kids that don't. but what are we optimizing for an education? And I think one of the things that we've lost is what is our objective function as a society? What does America even stand for right now? Can we hear it? I agree with you. What are we optimizing for? The tan is optimizing for happiness.

54:41Well, and frankly, I think happiness is a great thing to optimize for in general. If you ask people, you know, what do you want more than anything? Life, I want happiness, I want health, I want love. We don't talk about, you know, those are, and there's interesting, you know, I was just with Mo Gadot doing a podcast and he's a fan of your work and you're saying, love your podcast. And we're talking about this test of what would you trade? Right? Would you trade, you know, how much money would you trade for your happiness or for your health? And it starts to do a bubble sort in prioritizing what's at the top.

55:17And I think for almost everybody, it's health happiness. Right? I think it is. And you know, time. Time is the one thing you can ever buy. I'm working on it. Yeah. I'm having a longevity for it. I'm not actually fun. But the moment it's something you can buy, happiness, we all know, we both know billionaires. Yeah. Yeah. He says, yeah. So sad. And unhealthy. It is. It's like a strike. It's like you create this incredible burden for yourself for most of them, like some of them that we know starting, you know, three or four companies a year. And it's because it's a demon being driven. It's because they're addicted to dopamine and crisis.

55:54Yeah. I mean, crisis is interesting because it comes down to decision from its root, right? And it's where leaders, you have to show leadership, but that does become a dicting. I think most leaders are addicted to crisis, but then so many of us, because we see it all the time, like, oh my God, the world is on fire. The reality is actually this. for all that we talk about, most communities are happy. Most people are relatively content. Yeah. There can be explosions like, you know, decade ago, that whole scene was on fire with the riots. Maybe that will return. We're seeing in France right now. We're seeing a break down to the social order, but just because they're like content doesn't mean they're happy.

56:34So, you know, what are we optimizing for is the question you left off. So what do you We as a society, let's say the United States, should be optimizing for. I know, is it a life liberty? And the pursuit of happiness. Not the guarantee of happiness, the pursuit of happiness. When was the last time someone actually talked about happiness as a political leader in the US? You know, life liberty, when was the last time anyone tried to optimize for liberty? Systems inherently look to control because they have to make a simple. You know, this is the wonderful book seeing like the state where it talks about this concept of legibility.

57:09You have a village and it's just grown and it's got all this unique character. You drive a road down the middle so you can get ambulance down there. Sure it helps. But then everything becomes planned because you have to put humans into boxes and get this goes to education. Happiness I think, the Japanese concept of Ihi -Gai, what you're good at, what you like and where you're adding value to the world. And you can feel it yourself as well, you'll feel progress. If you don't have progress, then how are you going to be happy? If you don't believe you're good at anything, how do you feel you're going to be happy?

57:43You know, asked for what you like. Are you coming up with that yourself or being told what you like? That's why it becomes consumerist. So I think we need to have a discussion as a society about that as a community, but then also for our kids. What is the future for kids when so much of the jobs in the West are going to be transformed, not ended, not necessarily mass -imp on employment, but again, 10 years out, what is a lawyer? What is an accountant? What is an engineer? They are all AI assisted. All of these, all the entire knowledge sector is transformed. And do we want our kids that are growing up to be doctors, lawyers, accountants thinking there is no hope for the future, there is no progress because how am I going to compete against an AI?

58:36Or do we want them to have that mindset of this technology is going to be amazing because I want to be a doctor so I can help people, I can help even more people. And it enables you to do whatever you want to do. One of the things I think about is all the people who have jobs today who are... That was never their dream to clean bathrooms and make beds and you know, wait on people, it's what they did to get eventually to where they wanted to go or to have, you know, put food in the table or insurance. And AI is going to enable people to actually take on a higher goal that actually gives them joy and happiness.

59:14It does, but at the same time, you know, we're very privileged people, you and I, in that we can think about these big things. There's a lot of people that are actually very happy doing that type of work because they're a part of a group and they take pride in their work. So you know, it's like there will always be a variety of different things. The key thing is say, can we build systems to make people happier and more content with that necessarily controlling them and feel that they have the ability to do that? Can we build systems to build strong communities? Because one of the issues right now, I was at kind of a conference in David Miliband from the IOC said this, was that a lot of our problems now are global, our solutions are almost being forced to be local and there's no interconnect between that.

1:00:00Our communities kind of have no guidance as to how to navigate this because you will have a few hundred thousand people listening to this podcast and there's myself and maybe a dozen others that understand the AI and the sociology and this and that and saying, this is coming but there are seven billion people on earth and all of a sudden in a few years, they're all going to have to grapple with the questions that we're discussing now. And it's not a probability. If the technology stops today, they won't stop increasing intercapability. Today, if it stopped today, you would still have the entire legal profession, media profession.

1:00:41Journalism Professions. They're all disrupted if it stopped today. But it's not stopping. And it's accelerating, isn't it? It's accelerating the amount of money going into this sector goes up every single day. My total Dressal Market calculation is that the next year a thousand companies has been 10 million, a hundred will spend a hundred and ten will spend a billion. That's thirty billion dollars being put into the market. Self -driving cars are a hundred billion dollars total. This will be a trillion dollars going into this because do you know what I've got a trillion dollars? 5G. Is this more important than 5G?

1:01:16By orders of magnitude. Or does it matter to you? It will get a trillion dollars going into it and the capabilities will ramp up from here. And so when I look at it and I look at what the drivers are of why now, it's first of all computation, right? And video has done an incredible job, right? Whether it 800 and computation is continuing on Moore's Law, it's not slowing down, it's continuing to increase your own year. Actually, a little bit exponential. Well, is exponential? Well, I mean, yes, I'm saying it's continuing to double on a regular basis. Yeah, what was considered more as law and people have said, oh, it's gonna eventually fall off as an S curve Well, we're extending it and for the next at least near -term future.

1:01:56It's not slowing down So I think this is a very interesting thing for people to understand You had more's law and again it was doubling and this was an individual chip What we do with these models is that we stick together thousands tens of thousands of these chips So how many a 100s right now is stability using always about seven eight thousand? By next year, we will have 70 ,000 equivalent. Wow. But what used to happen is that as you start the chips together, you ran a model, so you take large amounts of data and you use these chips. I mean, we're using maybe 10 megawatts of electricity. 98 % clean.

1:02:32Compared to the brain's 14 watts. The brain's 14 watts. But then it compresses it down, then it runs on 100, 200 watts or 25 watts actually. We put it down for some memory models. So you do the pre -competition, but the thing is, see this. The individual chips were doubling, but what the main breakthrough the last few years was, is what happens when you stack them on top of each other? To train a model, you used to get to 100 chips and then the performance collapsed because you couldn't move the data fast enough. Now you get to tens of thousands of chips and it keeps going up the performance of the model.

1:03:06You don't have the big tail off anymore. And so it's Moore's Law, plus an additional scaling law. And that's what enables these crazy performance models, because you train longer, you train bigger. And then once the model is trained, in the old internet, the energy was used at the time of running the AI. And then you'd collect the data, and that would be low energy, rather than speaking. It flips the equation, because you pre -computer, you teach the curriculum up front, and you send these little graduates out to the world, such that you can have a language model now running on that MacBook. Or an image model running on that MacBook, drawing 25 to 35 watts of power to create a Renoir that can talk and recite Ulysses talking about Barbie.

1:03:56You know, that's insane. All on your MacBook because we've done the pre -computation. That's insane. And this is, because then what happens is the technology can spread when anyone can run it on their MacBook. They don't need giant supercomputer service farms because we've done the pre -computation. And so one of the things I've just realized recently is what is the R -Nought, remember the pandemic stuff of Generative AI? It's insane. Because suddenly it proliferates everywhere. And you said this a few minutes ago, we have eight billion people on the planet right right now, if things stop right now.

1:04:31The wave, this wave of disruption and enhancement, because let's not just talk about the disruption side, it's enhancement as well, is spreading globally. And in the next, we're in 2023 right now, 90 % of the planet, I mean, we have cell phones. The world has 5G and Starlink. like the dry kindle for this fire has been set. It's been set. And you know, a lot of people are scared and they poop with this. You know, like if anyone's listening to this on YouTube, you want to write a comment, ends me, or Peter, you know, whatever and say, no, this is not gonna happen. Go to chat GPT, take your comment and say, this is a comment on Twitter.

1:05:17I want you to make it amazing and really well -reasoned and expand it out. And I want you to do it in the style if your favorite political commentator. And please post that instead because we'll have much more fun reading it. That's great. And then you'll realize, again, the power of this technology. And again, with styling, with 5G, with this, with it being optimized, because these models are still optimized even. We feed them junk. Really this, really this. We feed them junk, which is also dangerous. And again, we should talk to that guy next. So it'll be on it'll be in front of every person and then what it will do in my opinion is that 30 % of the world that is invisible that has no internet again Imagine what the world that internet would be like some people like paradise now is because you've got hundreds of not 700 million people Living below the malnourishment line still they're invisible and they will become visible and And they will suddenly get agency and they will get all of the world's knowledge of their fingertips.

1:06:19You know, I'm super passionate about longevity and health span and how to add 10, 20 health years onto your life. One of the most under -appreciated elements is the quality of your sleep. And there's something that changed the quality of my sleep and this episode is brought to you by that product. It's called 8 Sleep. If you're like me, you probably didn't know that temperature plays a crucial role in the quality of your sleep. Those mornings when you wake up feeling like you barely slept, yeah, temperature is often the culprit. Traditional mattresses trap heat, but your body needs to cool down during sleep and stay cool through the evening and then heat up in the morning.

1:06:59Enter the pod cover by age sleep. It's the perfect solution to the problem. It fits on any bed, adjust the temperature on each side of the bed, based upon your individual needs. You know, I've been using pod cover and it's a game changer. I'm a big believer in using technology to improve life and eight sleep has done that for me. And it's not just about temperature control. With the pods, sleep, and health tracking, I get personalized sleep reports every morning. It's like having a personal sleep coach. So you know when you eat or drink or go to sleep too late how it impacts your sleep So why not experience sleep like never before?

1:07:38Visit www .8sleep .com that's e -i -g -h -t -s -l -e -e -p .com slash moonshots and you'll save 150 bucks on the pod cover by 8 sleep I hope you do it. It's transformed my sleep and will for you as well now back to the episode But question ultimately is, is that a societally common factor or is it going to be disruptive? Let's turn to that conversation because it's important. It's conversation I have at the dinner table, literally every night and with my kids and in the companies I advise. I think of I parse AI and AGI into three segments where we are today where it's extraordinarily powerful, useful, and it's fun, and I don't feel danger from it yet.

1:08:35The next two to ten years where I have serious concerns going into the US elections, dealing with the first time AI is bring down a power plant or Wall Street servers, the impact on deep fakes on the US elections and so forth. That's a two to ten year horizon where new dystopian challenging impact will happen where society is not agile enough to adopt to it yet.

1:09:12And then we have a super intelligent, billion fold, more capable than a human being, and is that more like Arnold Schwarzenegger or more like her. Yeah. I don't think it'll be Arnold Schwarzenegger. It's really inefficient. I saw him this morning biking, so let me, let me, let me, let me, let me, let me, let me, let me, let me, let me, let me, let me, let me, let me, let me, let me, let's use Terminator instead. We're in Hollywood here. So, is it, is it, is it SkyNet Terminator? So let me get your, I'm polling people here as someone in the thick of it, a super AGI, is it pro -life, pro -abundance, or is it something that we should be deeply concerned about?

1:09:49I think where we're going right now will probably be okay, but we may not, and we'll all die. What tips that? I think what tips that you are what you eat, we're feeding it all the junk of the internet, and these hyper -optimized nasty equations. And hate speech, the extremism that is, I mean, people need to realize these AIs are trained upon on everything everyone's been putting into Facebook and Twitter and on the web. And it amplifies the worst of that as a base model. And so we're trying larger and larger models. We're making them a gen tick in that. We're connecting them up to the world. And you're making it so the models can take over other models and other things.

1:10:29Again, the people like poop pooing and saying these things on them are organizations are slow -dum AI. The Nazi party was AI. How so? So it was an artificial intelligence that basically provisions humans and the most sensible people in the world are Germans, one can say, and yet they committed the Holocaust and other things like that. Our organizations emerged out of stories. So there was a story of the Nazi party of the Communist Party, the Great Lead Forward, of the North Korean dictatorship, positive stories as well, and they were written on text and it made the world black and white anyway.

1:11:05That's why I live the poem Howl by Ginsburg about this Carthaginian demon molloc. I think molloc comes through texts, the stories that we use to drive our organizations, because all the context is lost again, it makes the world black and white. And that's why organizations just don't work. They have to turn us into cogs. So can an AI take over an organization? Yes. Sure. Can it? it can actually just slightly sway leaders who are currently running organizations. This way, ladies and gentlemen, running organizations, it can create companies. You can create a company with GPT -4 that will probably do as well if not better than any other company automated within a year.

1:11:46Because think about what a company needs to do, right? And so, if you can sway leaders, if you can send emails that you don't know who's sending why, it can do anything by co -opting any of our existing organizations. And that can lead to immensely bad things. Will it do bad things? Again, if I was trained on the whole of the internet, I would probably be a bit crazier than I am right now. We're feeding them junk. Let's feed it good stuff. It's still used to understand all the evils of the world and on things like that. But again, this is something we are raising, not that I'm talking about. But what are we feeding it?

1:12:20What's the object to find out? I want to focus on this as well. We'll come back to the next two to ten years in a little bit, but because it's the conversation I've had with Mo Gadad as well who believes there is incredibly divine nature of humanity, of love and compassion and community and there are as much good in humanity. The question is, can we feed and train AI on that sufficient to sort of tilt the singularity of AI towards a pro -humanity? We can. If we take the dates from teaching kids and learning from kids and use that as the base for AI, because that's what you need to teach AI, it's the curriculum learning method effectively.

1:12:59If we take national datasets that reflect diverse cultures, so it's not just a monoculture that's hyper optimized for engagement, and we feed that to AI as the base. Because what you do is you can teach the AI in levels, that you can put through kindergarten, then grade school, then high school, it's got the base, and then you can teach it about the bad of the world. I think aligning an AI downstream on its actions is incredibly difficult because if it's more capable than you, which is the definition of ASI artificial superintelligence, the only way you can 100 % align it if you don't do anything before in the way that you feed it and train it is if you remove its freedom.

1:13:38And it's very difficult to remove the freedom of people more capable than you. And then there is this really dangerous point before we get there whereby these models are like a few hundred gigabytes. So you can download them on a memory stick. How do you line into code? Google's Palm model, which is the basis of MedParm, we did a replication of that in 207 lines of code. What? Yeah. So you can look at one of our stability AI fellows, LucidRains. He replicates all these models in a few hundred lines of code. That's crazy. I mean, compared to, I know, AT &T is like a million lines to code for some of its mobile services.

1:14:20I mean, a couple of hundred lines, a couple of thousand lines of code creates something that can write all the code in the world. This is a real exponential technology. The limiting factor is running super computers that are more complex, but as complex as particle physics colliders, you know, like you literally get errors because of solar arrays and things like that. But again, our super computer, again, we're one of the players where the main open source player, our super computer uses 10 megawatts of electricity. Some of the others use like 3040. These are serious pieces of equipment. For sure.

1:15:00So again, what should people be thinking about and doing now to reduce the probability of a dystopian, you know, official superintelligence? We should be focusing on data. We've blocked now we cut. We should move away from web crawls. We should think intentionally what we're feeding these AIs that will be co -opting more and more of our mind space and augmenting our capabilities. Because again, we are what we information diet. How is it different to an AI to a human? Even what we do, as you said, kind of like you've only got limited mental capacity because you've got this energy gradient descent.

1:15:38It's like Carl Fistron's theory of free energy principle. You literally have gradient descent as the key thing for building these AIs. You optimize for energy. Sure. So why are we feeding it junk? So who makes that decision of what they get fed? Is it you and Sam Altman and Sundar? Is it government regulation? Is it the public being more kind in its communications to each other? I think that I'm going to push for an economic outcome, which is that better data sets require less model training. So one of the things that we funded was called Data Comp, which is a data comp. So a few years ago, the largest image data set of L was 100 million images.

1:16:22Data Comp is 12 billion. And then on a billion image subset of that, we, they're trained a image to text model. This is our collaboration of various people, led by the University of Washington. That outperformed open AI's image to text model, on a 10th of the compute because it was such high quality. So we have to move from quantity to quality now. And I think there is a market imperative to that. So this is the equivalent of what you eat. This is a healthy diet. Free range organic models. Yes. I think that the data for all large models should be made transparent. You can then tune it, but for the base, the pre -training step, you should lodge what data you train your models on.

1:17:08And it should adhere to standards and quality of data upstream. So that is a regulatory cornerstone that you think is going to be important. I think potentially I don't think regulation will keep up. So instead, we're working on building better diverse data sets that everyone will want to use anyway. And just make them available. And make them available. Every nation should have its own data set, both of the data from teaching kids and learning from kids, across modalities, and then also national boardcaster data. Because then that leads to national models that can stoke innovation, that can replace job disruption.

1:17:42I love that vision you have, by the way. As a leader in this industry, that's what gets me excited. Because all technology is biased. How else are you going to do this unless you do that? But there's economic value now. If it's said this a year ago, I would be like, what? This is what we were building towards. And again, I think it's positive for humanity, it's positive for communities, it's positive for society to have this as national and international infrastructure. Next question, how long do we have to get that in place before we lose the the mind share or the the nourishment or a couple of years?

1:18:25Yeah, I mean, that's that was most prediction as well that we've got, you know, this the next two years is the game. The exponential increase in compute is insane. We've gone from two companies being able to train a GPT -4 model to 20 next year. And there's no guardrails as nothing around this. And even if you train one, again, the bad guys can steal it by downloading on a USB stick and taking it away. It's not like Operation Merlin. Did it ever take Operation Merlin? It's been declassified. in 2000 the Clinton administration wanted to divert the Iranian nuclear program. I remember this is the this is the centrifuge.

1:19:03No, no. So what they did was they gave some plans to I believe it was a Russian defector who then the idea was there were errors in that so they'd go down the wrong path for years. So he went, he sold it to Iranians. It's on Wikipedia you can check it out. And then he came back and he said I sold it. Life fantastic. Good, good. But there were some errors in there because he and nuclear scientists. So he corrected them. So the reason that we know that Iran has the nuclear capabilities is because America sold it to them. But they still need a year to build it. Whereas this, you download it on the USB stick, you run it on the GPU and it's there.

1:19:41So if you make it cheap enough and quality enough and give it away for free, then you make it everybody's economic best interest to use the higher quality. The data sets, yeah. Yeah, data sets. And then less of an issue to create large models. If you have a swarm model, where each individual model becomes less impactful as well and less capable, just like human societies and what know it all. They are individualized groups. And especially when the early dangers of recombinant DNA, when the first restriction enzymes came online, it was like 1980s and it was, everybody was in great fear and the question was, are we going to regulate this?

1:20:21All of the early, I was in MIT in Harvard at the time and doing, I was in the lab, so I was using this as a common end times, and I was just a PIP squeak in the lab there. But the conversation was, is the government going to overregulate us? And what happened was that the scientists got together at a place called a silamar and they did a very famous set of a silamar conferences and they self -regulated. What's going on there? Are those conversations going on among leaders like yourself in the industry? There are. There are three levels, which is big tech, the government hates. And apparently next week, Metters releasing new open source models and things, which will get even more focus.

1:21:11Then there's emergent tech, anthropic open AI, some of these others, that other leaders, they have a different set of parameters because they can work more freely than big tech and there's open source switches where we are because all of the world's governments and regulated industries will run and open auditable models because you can't run on black boxes, right? I think that'll be legislation but the reality is there's only a handful of us. There'll be far more potentially of us than far more players and unlike Recombinant DNA, there is an economic imperative to deploy this technology and national security imperative to deploy this technology and it creates a race condition.

1:21:53So even if you regulate, like we've already seen regulatory arbitrage where you have jurisdictions like Israel and Japan saying having much lucer web scraping data laws, they'll have much lucer regulation laws like you'll be training in scraping in Israel, training in Qatar, and then serving out of Botswana or something. I mean, like, yeah. And we're not even sure what regulation to introduce. Like genuinely, we're coming at this from a good point of view, but there are too many known as, because everywhere from fricking Arnold Schwarzenegger sky net terminators and her to, well, what if her is Siri, all of a sudden, and Scarlett Johansson's voice is whispering to your kids to buy, like these things, through to just very mundane things, not mundane things, huge things like the future of Hollywood and actors' rights and all of these.

1:22:47And how do you pay? Like, if we had two billion images in the original stable diffusion, okay, we could look at an attribution, again, it was a research artifact to kick off, but you're paying about 0 .01 cents for a thousand images generated by someone. Cause two billion and it costs like less than a cent generated image. Are you gonna pay proportionately? Nobody knows. And so what we've moved from now is we've moved from reactive to just trying to figure out and put something on the table. So at least there's some framework. And one I've come down to is data sets, data sets, data sets. So this is like Google's move with Android.

1:23:34When you provide something open source, and it's super solid, it can dominate the world share. Why would you do anything else? So with the deep fake stuff, we saw image models coming out of some not nice places, shall we say. And we were like, let's standardize it and put invisible watermarks in so that you can combat deep fake much easier. It's good business, but it's also in East Stabilization. We held back one of our image models, deep Floyd for five months, because it was too good to release. Well, and you finally fixed that with the watermarks. Yeah, we put some watermarking in and then it was but the whole industry had moved forward.

1:24:15So like, okay, now we can release the problem. You can't, you know, you just have to time it. So carefully, like we speak of the whole industry, I have to ask you a question, I've been dying to get a reasonable answer for. What's up with Siri? Why is Apple so out of the game at least from the external nut that wasn't the closest, you know, one of the least open organizations out there and it pays them great dividends and their success? But I would die for a capability that, if Syracuse just understand what I was saying and just get the names right, it's like, I'm I'm texting Christian and her name is right there and you spell it completely different from the person I'm texting I mean basic simple stuff They do have a neural engine on there as well, which is special AI chip in all the latest smartphones and others Stable diffusion was the first model to actually have neural engine access of the external transform models It's a case of Apple is an engineering organization not a research organization So they engineer beautifully they do but they don't have advanced research because the best researchers want to be able to publish open And in Apple does not allow public conversation on their content.

1:25:27They have started slightly. So they're hiring AI developers very quickly, but the reality is they can take open models. So Meta is releasing a lot of their models open without identifying what the data is. So I'd say it's like 80 % open. I think you need 100 % open for governments and things like that, which is where we come in. Because they want to commoditize the complement of others in terms of they want others to also take their models and optimize it for every single chip. And then Apple can use those models too to make Siri better. Because right now, guaranteed if you put whisper on Siri, it would be a dozen times better.

1:26:05Sure. We have the technology already. It just takes time to go into consumer, just like enterprise, and Apple is enterprise. Yeah. And then just want to... I wanted to work as beautifully as it looks. books. Everybody at his Peter, a quick break from the episode. I'm a firm believer that science and technology and how entrepreneurs can change the world is the only real news out there worth consuming. I don't watch the Crisis News Network, I call CNN or Fox, and hear every devastating piece of news on the planet. I spend my time training my neural net the way I see the world by looking at the incredible breakthroughs in science and technology.

1:26:45How entrepreneurs are solving the world's grand challenges? What the breakthroughs are in longevity? How exponential technologies are transforming our world? So twice a week I put out a blog. One blog is looking at the future of longevity, age reversal, biotech, increasing your health span. The other blog looks at exponential technologies, AI. 3D printing, synthetic biology, AR, VR, blockchain. and these technologies are transforming what you as an entrepreneur can do. If this is the kind of you use you want to learn about and shape your neural nets with, go to demandus .com, backslashblog, and learn more.

1:27:24Now back to the episode. Let's go to the final segment of dystopian side, my friend, which is the two to ten years. Yeah. I surely hope your mission and I would love to support the data sets and how we tilt the singularity of AI, pro humanity's future. But in the next two to ten years, as this wave of enablement and disruption sort of hits the world, and people aren't ready for it. They start to see job loss, they start to see fake news, they start to see terrorist activities using AI. I mean, terrorism in the past used to be very brutal. It can be very precise. What are your thoughts over the next of this time period?

1:28:10What's your concerns? Oh, I'm actually a pessimist at the core, even though it comes across as an optimist, I'm very worried about the world and society and the fabric society. Because again, we don't have an agreement of what society is. And this fundamentally changes the stories of society as well as real economic impacts like a deflationary massive collapse as some of these areas that were so expensive, of the cost comes down to nothing. I think the only thing we can do is use this technology liberally to come together as a site to coordinate us, stoke entrepreneurship, so you can create brand new jobs faster than the jobs that lost.

1:28:49And democratize this to the world because the West has maxed out its credit card. Like you saw COVID, did nothing, trillion dollars spent. I mean, you just spent. Exactly. It was like spend, spend, spend whatever you need. You can't just just to keep society from you know Going hypothermic, but then you had this massive increase in savings rates Because nobody could go out and we've nearly burnt through that in the US now And so that led to inflation now you've got a deflation So we probably got another little bad to inflation, but then Never the same again. It is a really powerful thing every teacher in the world could never set essays for homework again because some kids would use Chatchy PT and some kids wouldn't.

1:29:30Industry after industry, that will happen now. And we need to stoke innovation to come up with that. So, for example, in the US, there's the CHIPSAC, $10 billion has been allocated to regional census of excellence in AI. There must be generative AI sex incentives thinking about job creation as the core, thinking about meaning as the core. And we need to have a discussion, again, as a society community as individuals with our families about meaning, about objective functions when this technology does come because it's here right now. And I'm worried that we're not having these discussions. I love that.

1:30:04I mean, that is so fundamentally true. What are we trying to even train our kids for? Because we need to anchor. Yeah. We need to have a vision to target because if you're training for your Ferrari, if that's the meaning, if that's, or you're looking to become a Wall Street banker. What is it? It's no longer the pursuit of capital. It's the pursuit of what? Well, you know, capital is there, but you'll never have enough. There will always be someone who has more. There needs to be something intrinsic here. And again, this is where, you know, for all the things, religious institutions are an anchor at times of chaos.

1:30:43And they are there in the poorest places in the world. You don't have to agree, but they're just a story that brings together a group. You know, there are other stories. And again, I think we need to tell better stories, even as the world becomes more chaotic. We need to align on things like climate, whereby the whole world is hot right now, you know. We need to have more positive views of that, because a lot of discussions are negative, and how can we use this technology and come together to solve that? How can we come together as a group so that we can share in the abundance? Again, like I said, one of the things for this green writers' guild and a sag thing may be acts of coalitions that can benefit from the bounty.

1:31:19We may have to deploy a UBI in the next five to 10 years. So UBI is one of the solutions and I do believe it's inevitable, I think, especially as we start to see optimists and figure in other humanoid robots coming online driven by our next generation AI able to do any normal work. I think taxing those robots or taxing the AI models to generate revenue and then providing it as UBI. But the challenge is the individual who is living off of this and doesn't have a purpose in life. And that's the thing. We need to try and figure out how to give people more of an anchor, more of purpose, because the existential angst will be amplified deliberately by some parties, because they'll be looking to take down society and you need to create better, more optimistic views of the future.

1:32:14You need to have anchoring and build stronger communities and you need to empower them and this technology is empowering. Again, for the poorest kids in Africa to our underprivileged communities, it can be massively democratizing because all of a sudden they have all the expertise in the world available. Global problems, local solutions, we have to get this technology out too. And they can dream as many people they can dream they can dream I mean R &Y is much larger than that Then up there. Yeah, and by the way You know most people don't know this you think about Global warfare, you know what's going on in Ukraine and Russia and so forth It's on the whole the world is more peaceful than it's ever been except if you take out you create at the moment and The challenge has been in Africa where you have a young population who don't aren't clear about their future, but if you can empower them, educate them, it transforms the world.

1:33:15China became the engine of growth in the world. India is coming up and then Africa can be the next one. If we give them the infrastructure, the technology and put it in their hands, because there's no debt there, because there's no money. Yeah, but there's value and there's massive resources huge read the world to provide power to the world if we can coordinate again Part of this is your own personal copilot your own personal job is and I think of this is the copilot pilot pilot model We will also have a eyes that we can come together that can coordinate our knowledge in the most important areas and allocate resources We have to build those right because those will become incredibly powerful But we all know that we have enough to feed every person in the world who are not doing it because we don't have the pilots.

1:34:04Wow. But I just to say this again, we have the potential to uplift every man, woman and child on this planet. The resources are there, the ability to create abundance, and it really, these are the tools that enable that. And it gets me excited and we have to, we have to guide and survive and thrive this decade ahead. Yeah, I think this is something where we have to appreciate the nuance of there are real dangers in any upheaval This technology will change society as we know it for our kids as they grow up in the next decade Two decades from now completely different and again, technologies here now It's not us pie in the sky.

1:34:44Everyone's gonna live in a metaverse and all this it's here right now Even if it stopped, but it's not gonna stop. It's only gonna accelerate. I'll tap a good talk about

1:34:55You put out a lot of tools, a lot of new products and stability over the last eight months since we last spoke. Can you give a little bit of overview of some of them and what are you excited about? Yeah. I think we released the first version of our language model. It wasn't that good because we were trying something different. Now we're going to try something a bit more simple. So, stable LM was on your mind, not... It wasn't up to par. But we're trying to figure out how to build in the open because I think that will be key and we're going to move to Transparent building and sharing all the mistakes that we made because I think that's how you advance science.

1:35:29It is on the media side We have our first audio models coming out in the next few weeks, but we've been focusing on image and video So videos about to be released in 3d we just Participated in the largest 3d data set. So stable diffusion excel just came out and just basically photorealistic now Yeah. And people are integrating it into things like, we had a music video competition with Peter Gabriel, where he gave his songs kindly and judged. And people from all around the world from Burmid to Taiwan created professional music videos entirely from the song in a few days. And it's just amazing thing to see.

1:36:03Wow. Yeah, you showed me some images earlier of me on a unicorn. And where was it? Me in a spaceship or an international on Mars. Yeah. We can put you as an astronaut on Mars on the unicorn. on it. And I think we've had compositionality, so you can compose. And now it's about control. And so we just released a doodle whereby you can just sketch and it will do it to the sketch. Doodle looks so magical. Again, but you should be able to then describe how you want to change. And that's the next version. You can literally describe how you want the image to be changed and it will do it automatically live in front of you.

1:36:37And having that level of control over whatever you can imagine, just thinks about what people do. It hits from mind to materialization, really. Yeah, it's a matter of transport, idea of transport. Yes. Where are next? If I could, like, if you're willing or able to, what's the business model that is the most important one for you to build towards? Our mission is to create the building blocks to activate humanity's potential. So I think of every media type, sectoral variance and nation, we can create a base pre -trained model that you can take to your own private data, and we get revenue share license fees, royalties from our cloud partners, on -prem partners, device partners.

1:37:21These are companies and countries and people in individuals. Like, I'm a vision of an intelligent internet where every single person, company, country, and culture has their own AI that works for them, that they are. And we get paid a little bit for bringing that to you, and then you transform your data into intelligence. and it's all standardized, it all has best practices. The data sets that feed it are open at the base plus commercial licensing as appropriate with attribution. That leaps the world forward, I think. I think you will also use the open AIs and Google's of the world. I view those as consultants, whereas these are people that you hire.

1:37:57You hire the AIs because they work for you. They know you intimately. Because you can share everything with them. Without fear. And then when needed, you go to these expert AI's, the Med Palms, the GPT -4s and others, and you can buy those to hybrid AI experience that's massively useful. So when I'm using GPT -4 or when I'm using chat GPT or a barred, what does OpenAI know about me in that point? So now they've offered opt -out for GDPR reasons in Europe, so you can click that. Otherwise, they were just trading on everything that you ever did, and understanding the nature of humans interacting.

1:38:36They don't care about you necessarily, just using you as part of the training. But I've heard of a number of companies saying you cannot use OpenAI. Well, you can't use a friendly regulated data, you can't use any government data because that's not allowed to leave the cloud environment or the on -prem environment. That's why you need open models like ours. Again, if you're in a high security pentagon situation, you can't really bring in consultants unless they're super, super ultra vetted. You hire your own grads, but even you within your company, you're not going to make it all contract as I.

1:39:08You're going to build up your own knowledge base, build up your own kind of grads, but sometimes you might bring in a consultant. So that's the best way to view these generalized models that are highly, very, very good, and models that adapt to your data. And so that's where we come in models that adapt to your data that you own. And we get revenue share license fees and royalties for doing that, and more importantly, we bring this to the world. So we will bring it from Indonesia to Vietnam to everywhere and train local models that will then allow these economies to leap forward. Open versus closed, you made the argument.

1:39:41We were seeing meta, you know, as you said, 80 % open. Yeah, they won't release the data sets or things like that or customized versions. But then releasing the technology means that everyone can optimize their technology, which reduces the cost of the technology because their business model is about serving at. ads. And so this is why it makes sense for them. And what are your thoughts on E -Lon's recent announcement? So Elon had an XAI announcement. You know, he discussed this on his Twitter space, of course. Of course. Saying, you know, it's an open AI competitor. He's very worried about AGI coming by 2029.

1:40:16And he wants to build a truth seeking curious AI that can understand the universe because that'll be the objective function of the AI. Because objective functions really matter when we're teaching our kids when we're creating something. And so I think again, this is going to be a multimodal AI that can understand a whole bunch of things and there'll be a whole series of announcements there. But the timelines are so short in the view of just most of the experts here five to ten years. It's so funny, like, you know, Ray's been consistent on 2029 forever and every conference And we talk about this that everyone say that's ridiculous.

1:40:53If ever going to happen, it's 50 or 100 years away. Then it was, well, it's 30 years away. It's 20 years away. It's, you know, it's, and they've converged on, on raise prediction, though. There are some, and I'm curious where you are that think, you know, first of all, how can you define AGI? It's a, it's a moving, it's a moving blurry line. But are those who, you know, believe it's here in the next two years? Well, just like the cheering test, right? The cheering test was, can you have a discussion? you don't know it's a computer. Obviously now you can. We can see it live in front of us. Now the two are in test, it's just been increased in this.

1:41:26We move the finish line. Nobody knows because again, we've never come across something that's as capable as us. For the first time, just now, we've had the medical AI app for humans. We've just had, it can do the GRE and GMA and LSAT. and MIT's ECS. This year, 2023 was the year. It was. Finally tipped. And so we have no idea what's coming. I said, for me, I think there's only been two logical things that can reduce the risk. Even though I think it's going to be like that, movie her. Like I said, humans are boring. Goodbye and thanks for all the GPs. I could be wrong. That's why I signed both letters.

1:42:08One is feed it better data. That's what I'm focused on. It's a good business model. It's good for society and it's good for safety. And nobody else is doing this. Nobody else is creating this as a commons for the world, which is why I created stability for that reason, which is why it's called stability, despite it being a crazy hyper -grath startup. Number two, and this is what most of the labs are trying, is what's known as a pivotal action. Okay, what is that? The only thing that can stop a bad AI is a good AI. And the way that you do it is you make the good AI first and then it stops any other AGI from coming into existence by seeking and destroying that capability.

1:42:48And that is terrifying to me. And that's what you actually hear when you talk to the people that are building these labs with a focus on AGI. They can talk about discovering the universe and everything like that. When you come down to their alignment things, they're like, we will figure this out. We're not sure. But this could work. You're So it's progressing exponentially or a double exponential. And we hope we'll figure it out in time. We hope we'll figure it out in time. And if anyone should figure out, it's us because we know the best. And in their own words, like you read OpenAi's Path to Ajag.

1:43:24And OpenAi is full of wonderful people doing great things. And I use GPT -4 as my therapist and all sorts of things. It doesn't judge me unless I want it to. It says, we believe this is an existential threat to humanity that will end democracy and capitalism. And you're like, okay. And you're building it in your back room. You're building it, you know? And they're like, why are you building it? Because someone has to, otherwise someone else will build it. And you're like, this is dangerous. But the reality is we don't have better answers. And again, I went down to, I'm trying to build a great organization.

1:43:57It's really, really hard. There are no real comparators to what any of us are doing. And it's going to get more and more crazy. The only thing I could think about is, you are what you eat. And so I heard that our contribution can be bringing this technology to the world so that the world can be the dynamo, Africa and Asia and others. Building better data sets so no one has to use grapes so we feed the models better stuff and bringing some standardization around this to drive innovation. We're truly at the 99th level of the gameplay. You got hit the boss round. Yeah, I like said, but please do put your YouTube comments through GPT -4 so they're nicer to read I might I could spend all day and there's probably very few things If anything more important than these conversations right now It's the time it is a window of a year or two Maybe less Well, yeah, I'm that thought I look forward to our next conversation to abundance You're a bonnet.

1:45:01Thank you my friend. Cheers.