Transcription
An AI will be able to replace your job by reading every single mistake you've made, every single outfit you've done, and creating a virtual twin of you. If you can do it on the other side of a screen, the AI is going to be better than you in a few years. But what doesn't change is your connection to other people, how you identify yourself. It is this very human thing of our identities are tied up with our jobs and other stuff.
There will be this backlash against AI coming next year, but those that use this technology to build stuff will out compete those that don't. And the important thing is >> Immad Moto is the AI pioneer who reshaped entire industries through AI. >> But now he's revealing how AI will evolve in the next 1,000 days and how you can survive what's coming. >> Most AI leaders are focused on let's build this technology. But when OpenAI says 200 billion of revenue, 100 billion of that is AI agents that replace humans. But what does it do to society? And they're like, well, it's not my responsibility.
What really matters is the AI that is next to your kids that advises the governments that manage your healthcare. Who's that AI working for? That needs to be distributed. It needs to be sovereign. Otherwise, we're going to end up in a really weird world. >> What does this actually do to the future of humanity? >> It's hard because so many things are going to change when humans are the dumbest people [music] on the team. As we go past 1,000 days, the next 10 years, the cost of cognition is going to zero. It may or may not kill us all. And there are ways that maybe we can help with that. But right now, what's certain is it will start displacing more and more of the economy with real human consequences. Every person, every [music] community, and every country, we've got to focus on Emad. Welcome to the show. Is great to see you here. We were actually meant to meet at Token 2049 in Singapore about a month ago now. We couldn't make it happen, but we're here virtual, and it's great to see you here. So thank you so much for your time.
>> Yeah, it's a pleasure to connect finally even if it is virtually hopefully next time in person. >> Let's do it. Well, I would love to kick off AI. It has been changing so quickly and I think it would be helpful for our audience to get a better sense of kind of what's happening right now because I don't know about you, but it's moving so quickly and I'm kind of terrified. [laughter]
>> I think uh it's right to be a bit terrified, but also hopeful. This is like the most interesting time. I think what you've got is you've had the first generation of this new generative AI where you typed in prompts and it created images or text instantly. And it was like a really talented buddy that you could bounce stuff off, but it didn't have any of this memory long-term thinking ability to do really useful economic stuff. Now what's happening is we're at this inflection point with models that can reason, that can think, that can act proactively that are just hitting the market now and over the next year they'll full move into full agentic systems. And this is the next takeoff point that we'll see.
>> It's all happening so quickly and we're going to chat so much more about this later on, but let's go back to the early beginnings. You started off as a hedge funds manager and then decided to launch your own company. go on this crazy ride of building your own company and obviously something which went super viral stability AI stable diffusion what was that kind of eureka moment I mean I remember when this was like kicking off I think in around 2019 2020 and it popped like just like that
>> yeah it was actually 2022 again AI timelines seem a lot longer than they are um in 2020 I was leading an UN AI initiative against CO 19 and we were promised a lot of technology by these big companies and we weren't given it cuz it was too dangerous, etc. So, in 2021, I was thinking like, this isn't right, you know, like who's going to build the AI that does all of these things. And I was like, well, no one's going to build it. Let's build it ourselves. And there was language stuff going on, but it wasn't quite there. In image, I saw that this was the big breakthrough. So, we spun up some supercomputers. We built a community like Luther AI and Lion. And then we started building models. And the big breakthrough was in the summer of 2022. stable diffusion which was the first real text to image model where you typed in words and it created actually really good output and we managed to make it so that it could run on basically any graphics card. So there was this big democratizing moment and we released it in August of 2022 and it ended up with hundreds of millions of downloads of the model and billions of images created from it.
>> 300 million downloads. I mean this is crazy. This is like what every founder dreams of, right? What was that first viral moment for you?
>> So you that was the first model. We followed it up with audio and video and 3D and kicked off the whole generative media thing. But it was a crazy time. Like this was before chat GPT in November of 2022. It was the first big generative AI breakthrough. Like in the first week when we released the hosted version of this, we had 3 million users on our platform, etc. And it was crazy because we've never seen something like this where you could have this tiny file of a few gigabytes that almost anyone could build on and a whole community built around it, making it into things like, you know, the thing where you uploaded your face and it made you into an astronaut. That was this technology, you know, editing technology, all of this. And it was crazy being in the middle of that. It was a real mastrom where everyone wanted to talk to you and everything was interesting. And then you got to the next stage which was like, oh wait, this is changing the way that creativity works. This is changing so many industries. How do we do this responsibly? Etc. Like things you don't even think about when you're just doing research and experimentation thinking how can you take words and turn them into images, you know?
>> And again, this was just 3 years ago, right? And it's crazy just how fast the internet and AI has moved beyond since then.
>> Yeah. And again, it's 3 years since then, 3 years since chat GBT. 2 years ago, today actually, I think, we released stable video diffusion, which was a state-of-the-art Nvidia models, and it was like taking a stat picture and then it'd be like slow movements and things like that. And now you've got Sora 2 based on that same technology, etc., which you just can't tell from reality. And I think again, this has been that takeoff period of the creation phase. And now AI models can create things from prompts really well, but the next stage is now they will tie them all together into economically productive stuff. You talk around we have a,000 days left until AI completely changes basically the future work forever. We're going to get on to that. But when a company reaches 300 million downloads, most founders decide to stay on. they say they decide to go and launch new companies or even build a product even bigger and better. You decided to choose the complete opposite route and actually leave and go and start your new project. What was that kind of thinking and inflection point why you wanted to make a bigger impact on the ecosystem?
>> Yeah. So, the start of last year, I looked at where things were going and I was like, the media side's interesting and we'd built this suite and it's going to get even better. And the round about 2026, I said it would be Hollywood level, which I think is matching up with that. Uh, in the start of last year, we raised 100 million in uncapped safes [laughter] um from Intel and others, which is a crazy thing. There's no valuation even. and we were going to do a bigger round and we got to about 5.3 million a month in revenue on 8 million of spend despite doing these state-of-the-art models which again was pretty great in terms of the overall thing but I looked at everything I was like this isn't really what I want to do um what really matters is that the AI is going to get to human capability and then start displacing the economy and the important thing is the AI that is next to your kids the AI that advises the governments the AI that manages your healthare that needs to be distributed. It needs to be sovereign. Otherwise, we're going to end up in a really weird world. And that's why I was like, we got to build this distributed AI system that is available to everyone. Make sure no one's left behind and think about new ways to do the whole economy itself because so many things are going to change when humans are the weakest part in the cognitive link. You know, the dumbest people on the team. And that was a crazy thing. And then I took 6 months off and then now we're doing the new company intelligent internet.
>> So intelligent internet I want to give you this time to explain what Elashi is as I don't know about you but my friends my family um at some point our kids generation when we have them they are going to enter a completely different workforce. What is the purpose of intelligent internet and do we need you know another AI company out there?
>> So the purpose of intelligent internet was to build the AI that is closest to us civic AI and do it in a way that's distributed and owned by everyone as essential infrastructure. I just had a baby girl earlier this year and her best friend will be an AI. Her first love may well be an AI. But who's that AI working for? You know, like I want to be able to use chat GPT to teach her but not have chat GPT teacher cuz you can get very strange outcomes when you have private companies doing this. It would be like exposing her to Fox News or China, national TV or Alazer early. Like I want to be able to control that and I want her to be able to control the AI that's closest to her. So in building the intelligent internet, we were like let's build really amazing agentic stacks for education, health, finance, government. this civic AI where the data should be transparent, the decision-m processes should be transparent and then make that available to the world in a way that it can be owned by everyone. So, we've built the world's best open healthcare models that are better than human doctors that work on laptops. We built the world's best agent III agent that outperforms Codeex and Claude and Cognition and others and will be open sourcing in a few weeks. Best data sets and more. And we want to iterate that stack so that it's available for every country, every culture, every sector next year, tied again together in a way that's very distributed, building off the concepts of Bitcoin, Ethereum, and others.
>> Wow. Next year.
>> Yeah. Like you can use the tools today, the individual building blocks, but really we need to build that full stack for a government. So we announced a project called Sage, the sovereign AI governance engine in Saudi Arabia, big conference FI with the backing of multiple nations, which is an AI that helps governments build policy. And the moment there's a breakthrough in AI, it automatically updates all policy and all research and it can generate slides, presentations, simulations, and more. And we're going to open source that and make that available to every government in Q1 of next year. On the healthcare side, we have the best model. But what you want is an open system that can run hospitals that you can install on your own data that can do a second opinion by this time next year of every medical decision in every country that will save lives. For education, you need to think what does the AI first education system look like? And again, we believe that should be open and customizable. So that's why like we spent the last year building all these building blocks achieving state-of-the-art cuz no one else was really doing that. Everyone else is stuck in this thing of well I'm going to build the best healthcare model and I'm going to charge access to it. You know they're like that leads to some weird outcomes. You know I'm going to build the best government model and guide governments with it. But again that should be an open model. It shouldn't be a closed model. Do I want a black box from Anthropic running my government and doing government policy? Probably not, right? Do I want again the AI that's closest to my kids being a meta AI? No, I want them to have their own Jarvis that's looking out for their flourishing. And the same with me as an individual. So taking that stack took a little bit of time to build, but now all the building blocks are nearly there. And then that feeds into the work and research we've been doing on the economic side because we think to guide through what's coming this big disruption of the next thousand days when these AIs all the building blocks come together and they can replace the cognitive workforce and then the physical workforce once robots get going. You need to put an AI next to every person, every community, and every country that can guide them through what's coming and that they can trust because it's verifiable what these things are doing for the flourishing of individuals, community, and humanity.
I don't know about you, but it's just so a thousand days ago, right, Chad GPT now is like now it's completely changed. Now all of a sudden, you know, what you're talking about is is happening. You mentioned around the uh the healthcare model. So how would that work um in principle?
>> So right now um human doctors on this benchmark open AI did called healthbench scored 20%. Chad GPT scores 50%. And our 8 billion parameter model that works on a Raspberry Pi scores 48%. So this works on anything any phone any Raspberry Pi but it's not good enough still. You need to be 10 times better you know. So next year we will have an app that any hospital can download, any doctor can download in every language that does a second opinion on every medical diagnosis. In fact in a few years we believe it should be malpractice not to have a second opinion with AI. Years from now you will have AI making the medical opinion but that depends on society right just like with self-driving cars. If every self-driving car, if every car in America was the same safety profile as a Whimo, you would save 40,000 lives a year and a trillion dollars in secondary costs. But it'll probably be a decade or more before self-driving cars are mandated, you know. But first of all, you need the assistance, then you need the replacement. So we're focusing on the assistance first which is second diagnosis making this freely available to everyone packaging it with solar panels like can you get a $100 doctor that you can give to any village in the world and I think you can by next year and so this is the power of this technology again as you think of it as infrastructure as a human right.
What I find also really interesting about the product is I heard you say in some of your earlier episodes uh interviews, you can use this product without a internet connection.
>> Yeah, I think this is kind of the thing like if you look at the state-of-the-art model last year, right now you can get equivalent performance on your MacBook for an edge model. And so everyone was like AI is going to go like this exponential and you're going to need billions of dollars of compute. The Chinese models, our models and others have shown that actually the cost of intelligence is trending towards the cost of electricity. But then the defaults on that it going to be so immense. I call it cognitive colonialism, right? Like if your AI by default is looking out for open AI or looking out for Google, then it's going to be biased and again that's going to be closest to you. What goes inside these AIs, especially as they can go on the edge on existing infrastructure to everyone at almost no cost, is so important. Those defaults are so important. You know, you wouldn't invite someone in who you don't know what their background is to your house. But again, we're making it as close to us as possible cognitively. This Jarvis of the future. So that first layer, that control layer needs to be open and it needs to be verifiable. And again, that's what we're really focused on
>> for sure. I think um are we going to see two scenarios where and I've heard you talk a lot around this type of model and you know with the coins and blockchain and so on. Are we going to see two models or two scenarios where a lot of these AI models are owned by these corporations and then you have the other side which is um sovereign owned. Is that kind of where we're going to go?
>> Yeah. So, we had three different scenarios in the book. the last economy we were like you know this great fragmentation that occurs where everyone countries got their own AIs and there's no interoperability this kind of version where the big companies have this colog of colonialism and control everything so when it says beer it says Budlight cuz Budweis has paid them a bunch of money etc and this final thing is this symbiosis where there's sovereign AI on an individual community and country level and again I think the interface layer is the most important part of that like chachi PT anthropic they're wonderful technology ies, but I don't want them owning my identity in storage. Like already if you do anything with chat GPT today, even if it's that temporary hidden version, they store every single one of your log files and the New York Times can actually view those in discovery due to their lawsuit that they've got with them. You know, like people are telling it really personal stuff and it's not going anywhere. In fact, in some cases, it might even be used as training data. So, I'd rather prefer to use these other ones as API inputs. I don't think they should control that layer next to the user, just like a TV broadcaster, you would want to have a control element, etc. And I think again, we can set those defaults and do that if we build really good chat other experiences individually and give them free to people. So, for example, in a few weeks, we're going to release our version of chat GPT, but it's just like chat GPT except for you can use every single chat service or model at cost because we're like that makes sense and you can use our agent at the same time because we're like that makes sense. Like again that interface layer is important and you can either run it hosted or you can install it yourself and have the same functionality because again that's what's important that interface layer.
Wow, that would be incredible. I think it's uh it's just so hard to fathom you know how far this whole category has evolved and it's only going to keep changing. um a lot of conversations are happening right now around what AI models get wrong. What do you think is what's your take on that? What are what's the most challenging aspects of these models right now?
So there was this whole thing about hallucinations particularly for other stuff but you know humans make a lot of errors like in medical diagnosis the error rate is like 20%. Which is unbelievable you know but again AI makes a lot less and if you look at GPT5 it's gone from an average 14% error rate to 2%. So you're getting lower and lower on the error rates. They're making less and less mistakes. There are more and more ways of them having memory, continuous learning, and more. And again, this is the next takeoff point that we've got here. This happens at the same time as the cost of intelligence per unit drops dramatically. Like GPT3 was $600 per million tokens. It's like 700,000 tokens for a million words. GPT4 was $100. GPT4 was $40. GT5 is $10 and the latest Chinese models are 50 cents per million tokens per million words. The average person only speaks 10 million a year. You know, that's like five bucks a year. So, you can apply a lot more really highquality tokens and intelligence to the same problem, which means you'll make less and less errors. So, a lot of the issues that people have flagged even from 6 months ago, a year ago, have just disappeared now. Like, they've been solved.
It's um and it's only going to keep changing, right? And I think this is uh you know a good segue into your latest project, The Last Economy, one of well your your latest book, right? Um which everybody should check out. And it basically you know articulates you know where this whole category is going. And I think an important conversation around today's episode is around what does this actually do to the future of humanity? What does it do for our jobs? You know, I'm 27 now. I've got family, brothers, sisters coming into the workforce now. And what they've been studying for the last four, five years at university, it's going to look completely different. Some of it's not relevant anymore. What's going to happen to those people?
>> Yeah. Like entire industries, if you can do it on the other side of a screen with keyboard, video, and mouse, the AI will be able to do it better next year. Right now, it's the graduate workforce that's challenging. Like again, if you've got anyone who's just graduated that you know where listeners do, it's hard. It's not a good job market at all despite the S&P and other things at highs, right? And we've seen that there's a paper by Eric Benson and Stanford where they've shown graduate labor is finding it hard to find jobs particularly in cognitive fields because the people with a few years experience now with AI are much better than they used to be because AI is an uplifter. But then why would you deal with the new graduates or undergraduates? You know, it doesn't make sense. And that's just going to get worse and worse because in a year maximum two, an AI will be able to replace your cognitive job by reading everything you've created in the company, every single mistake you've made, every single output you've done, and creating a virtual twin of you that you can talk to on a Zoom or WhatsApp, you can give a call to, you can't even tell it's not a human on the other side. And that will be the case for all cognitive jobs, I think, within the next three years. Now, doesn't mean you'll get fired. It just means you won't get hired initially, right? If there's a downturn, then people start to lay off and they just don't rehire because they get AI instead. Cuz why would you need to? And there's whole categories of jobs which don't make sense. Like, you do computer science at university, you're not going to be able to build better apps than an AI in a few years. Why did you spend all that time learning to code when you could have just built stuff? You know, if you're training to like do customer service, like your customer services jobs are gone. Like there's so many jobs they're just going to go one after the other. Particularly those jobs where they are teaching you to be a machine, the machines will be able to be machines better.
>> It's just crazy, right? It's like um coders now. I think AI can outperform 99% of coders probably. Yeah, AI came like I they got gold medals in the recent coding competitions and on an ILO basis on code forces which is competitive coding, it's now basically top. So the AI that you can use today for $20 a month is the same as a top 10 coder. Now it's not that it will produce go to that quality always because again we have to build the infrastructure and scaffolding around it, but that's only a year away at most.
>> Wow. Um, so in 2023, I said it would be 2026. Like that's the year that this kicks off. And in 2028, I would expect there to be no more coding jobs as we know it. Like you'll still build stuff, but the language will not be Java or Haskell or Python. It will be English. Wow. Okay. So, custom service jobs basically gone, almost gone. Coding, um, probably a bunch of others. But what comes next? What do you think is that undisrupted market? uh public sector is going to be undisrupted like you know those it's not like those jobs are optimizing for efficiency right
>> yeah [laughter] >> in fact the interesting thing will be when you have entirely AI firms how you going to compete against them they don't sleep they learn from their mistakes they optimize they can proliferate their workers from one to a million with a single click of a button so more and more jobs will go to the public sector this is the same as 1929 post the great depression the great jobs programs, you know, the New Deal and other things like that. So that would be my expectation that you'll have COVID type mass payments, shall we say, but that's just not sustainable even as more and more of the economy becomes AI first companies or AI companies competing. This is Elon Musk's bet with macro hard, you know, his version of Microsoft. Those millions of GPUs in the data center are not to give people pretty gro and things like that. It's to create the accountants and lawyers and others of the future. Like Grock is already a better lawyer than most lawyers. A Grock that you can literally talk to on a Zoom that can read through your things like they will be the best lawyer in the world. Rock law within 2 years
>> is and there are similar companies already achieving those rights for lawyers accounting now finance.
>> Yeah [clears throat] for individual task the AI is better but the next stage was putting it together into workflows and having these reasoning models that could do tasks going from a few seconds now to up to 7 hours and in the next year or two they'll be able to do arbitrarily long tasks. So the whole process of making a video is lots of little tasks. And right now we can do individual scenes. You will have AI directors and producers and AIs that go out and negotiate IP rights and things like that. And you won't tell it's not a human doing these things to make entire feature films.
Who um who wins in this space? Um is it the founders of these companies? Because at the end of the day, people are going to become immensely wealthy, very powerful, and what happens to the rest of us?
>> Yeah. So, in the near term, it's those that embrace this technology because it's such a force multiplier. Like anyone who's listening to this that's tried to build a team or organization, what's the most difficult thing? It's people, right? You couldn't scale your talent properly. Now, you push a button, you get the talent. And if you can use the technology, right, you can do so much more. like you have companies with 20 people getting to hundreds of millions, billions of dollars of revenue now, right? And so I think it's the people in the short term, especially because the cost of each piece of intelligence is dropping by 10 times a year, 100 times a year. So again, if you take an AI that speaks as much as a human, 10 million tokens a year, shall we say, that would have been $6,000 worth of tokens in GPT3. Now next year it'll be 5 cents. You know the cost of making a movie like how much is a VO3 clip? The average movie scene length is 2.5 seconds it's like 10 cents. These costs are collapsing and the Chinese are open sourcing everything. So the cost of intelligence will approach the cost of electricity and it's how do you use that intelligence to solve problems because the AI models by themselves are like um Nim Talib the author of the black swan had this thing that really smart people become really dumb when they enter politics or run companies and things like so misalign incentives right so there's no skin in the game they're intellectual yet idiot AI systems are the same they don't care you give them a prompt the next second it's gone, right? They're just it's a new thing, a new thing, a new thing. When you apply these with skin in the game and you think of them as a leverage workforce, you'll be able to out compete everyone who doesn't and take on and beat other entities. Before again, you have fully AI firms, but then that is your measure of comparative advantage and success. And there's a real sweet spot now where anyone that really gets in and uses technology can do incredibly well and build up wealth. But the question is what's on the other side which is the really interesting thing as we go past a thousand days to the next 10 years.
>> What is on the other side?
>> I don't know. I [laughter] think you need to have a universal basic income combined with a universal basic AI where everyone's got this capability. We need to coordinate through it and we need to rethink things when capital doesn't need labor anymore. You know, like if you wanted to increase your capital, you had to involve labor in some ways. Now you just got to buy more GPUs, right?
>> Exactly. It's um it's terrifying. It's exciting, but it's terrifying. Mostly terrifying. Um
>> it's just changing so quickly. Well, you've got a Star Wars future of like these massive degradations and other things, zero sum. When you've got the Star Trek future of abundance, when the AI can do all the work and it can feed everyone, you know, like does humanity return to community? Does it return to network? Like, do we explore the universe? Like, we have enough resources to feed everyone and give everyone a nice life. What gets in the way? Our bad systems. the AI can make the system smarter but just like that foundation by Asimov the question is can we build the infrastructure to coordinate through this with the right incentive structure and so again that's what we're trying to do with the intelligent internet and things like stage for policy advice and then universal AI provision through II agent etc like make that a communal thing that can help us coordinate with good intention versus profit and engagement maximization
>> totally how would ads operate across each government policy maker.
>> So we're building this open system that can advise on policy. We announced like I said in Saudi this Sage system that does in policy dynamically. So you have a breakthrough in AI. It feeds through to everything living documents, simulations, websites, powerpoints. You can't have static PDF anymore. But then what we're doing is we have a concept that we'll be talking more about called national champions where for every country. We're setting up a national champion owned by the citizens and the local entities entirely that we're enabling to run this infrastructure to give free AI to the people and then bring it all together through this Bitcoin equivalent we have foundation coin cuz the compute of Malaysia is going to go from nothing to lots. the public sector civic AI compute of that can be a minor on the network because it'll be more and more compute and more and more network by helping people through giving them free aligned AI and that's what we saw the future should be like you need to fund this all somehow so fund it through monetary creation by helping people every coin sold goes to compute to help people be it Malaysia or cancer etc you know building off this bitcoin legacy have that as a coordination system where everyone builds this open stack whereby once you solve loneliness and the elderly in Detroit, it can go to Dhaka as well with an AI that calls them and sees how they are. Once you build a supercomputer for cancer and you do a coin cell for that, suddenly everyone going through their cancer journey gets a customized AI for them that helps them through their cancer journey and connects them with other people, you know, gives them the latest knowledge at their fingertips in every language. And we thought this is how you should build again an infrastructure. Um and we outlined more of that in our white paper. But we thought again we need to change how money and value is created to this network becomes more valuable the more people it helps. The more compute it stacks and uses to help people. And that's the feedback loop that we need because you can achieve huge amounts of good if you stack compute and use it in the right way as opposed to AI slot factories or whatever. Like again Samman said I'd rather yeah do I use 10 G of ed of compute for like chat GPT or cancer. Yeah let's try and make it so it goes towards cancer right and solving that. But we need to create the right incentive mechanism for that and I think digital currency is a great way to do that.
There are a whole bunch of AI leaders now um who've emerged in the last thousand days. Yourself, the Sam Alman's, Satia like and a whole bunch of others. Are you guys do you guys speak about what you're all doing together? Do you speak about impact? Do you have them on speed dial? I talked to some every so often. Most of the discussions were about is this AI going to kill us all? And then regulation to be honest like um nobody one of the reasons I left the company you know like again there was so dramatic and there was lots of espionage and other stuff cuz the stakes are high was just because I was like who's actually trying to build intentionally this structure and we didn't see anyone. So took the core team from stability with me, people who helped start it and again like let's just build this intentionally cuz part of it is building the system where everyone can give their input. So sage is the first version of that. It dynamically creates policy but anyone can give their input from experts through individuals and that's obviously the way that laws should be done, right? It's the way that policy should be done but we never had enough policy makers and systems to engage with that. So now I'm talking to a lot of them saying let's help have this transparent input policy creation engine and that'll be ready within a few months for everyone using a thousand black wells or so. So a really good amount of compute other than that people tend to be stuck in local maxima whereby it's just overwhelming like again if you look at Sam Alman he says intelligence will be the cost of electricity and he knows that you know you look at satis like this will transform entire industries and have the potential uplift but what we've lacked again is a single coordination mechanism that is credibly neutral. You know this is the very interesting thing about blockchains. Blockchains do well when they're credibly neutral. Can we build AI systems that are credibly neutral to help organize and guide us through what's coming? Other than that, it's this big thing is coming and we can see it. And even people like Gary Marcus and others that are bearish say AGI will come by 2040. You know, it's like bullish is one year, bearish is 10 years. Everyone says this stuff is going to come. But how do we coordinate? We need to build the coordination system. Other than that, if we just minimize it for maximum regret, what if it kills us all? That debate's a bit played out. It may or may not kill us all, and there are ways that maybe we can help with that. There's Ellie Isaac Kadowski's new book about that and things like that. But right now, what's certain is it's just going to get better and it will start displacing more and more of the economy very soon with real human consequences. And it'll happen at a pace not like the iPhone or the internet where you had to build a whole infrastructure. Today, you have a medical AI model that can outdose any doctor pretty much that works on a Raspberry Pi or a 12-year-old computer. It's not hooked up yet. It will be next year. And that can be a call center worker. That can be a tax accountant. The cost of cognition is going to zero.
is uh yeah I mean it's happening um and it's only going to get better right help how help our audience break down what's going to happen so we're nearly at the end of 25 so let's say end of 26 any end of 27 28 next thousand days what's going to happen on a yearly basis so end of 26 you will have probably an AI that for many cognitive jobs can create a digital twin of you on the other side of a keyboard video mouse and replace you like talented Mr. Ripley style, right? And it will do your job better than you can and it will have your mannerisms and everything like that. That technology is capable. You won't see mass firings unless you have an economic downturn because companies don't like to fire people, you know, like it's bad for morale and more than that. Like people do the job good enough. If you have a downturn though, people will start losing their jobs and on the flip side, 2027, they're not going to get rehired. So graduate entry- levelvel jobs dry up almost completely for cognitive sector. Then you start to see that displacement and then 27 you start to see the first robots that can do physical tasks better and they might cost $50,000 but $50,000 is $5 an hour. You know, by 2028, they'll probably get down towards $20,000. And by 2028, you'll definitely have the full stack of almost every job that can be done remotely, can be done by an AI better. And the cost of that will be probably a hundred bucks a year for that AI to do your job better. So, we have a,000 days to make as much of an impact as we can or people entering the work pools to get involved as much as they can in AI. So look, if you're a grad right now and you're finding it tough, what should you do? You should go and build stuff, right? Like use the replets, the II agents of the world, the sunnos, the loomers on audio and video to just build stuff. And that should be your resume. Your resume being I went to this university, I did this isn't good enough. People want to see what you build. And those that use this technology to build stuff will out compete those that don't. either in the workforce, they'll be the last to go, the first to be hired, or by launching their own things because suddenly you can scale businesses, you will have one person unicorns. The whole sales, marketing, distribution, social media will all be AI agents. And that's a 2027 thing, 2026 thing. So if you're first on that and you're solving real problems if you have skin in the game, you will out compete those that don't. Cuz right now most people are still used to GPT4. They only just got GPT5
>> and they're using it in this very very shallow way.
>> Um, if you were a if you're a graduate today or someone who's just lost their job because of AI, but they're super hungry to build, what is what are some of the opportunities out there? Okay, you've got the sales, you've got the marketing, but someone who's, you know, deep ingrained into this category. Where is some of those untapped opportunities?
>> So, it's improving it's reducing cost. It's improving revenue, right? And again, your CV should not be the static CV. You should go and use replet to say here are some problems and pain points within this industry that I'm in that I want to address. Let's build software for that. And you can do it just by talking to it and showing that demo to people as your CV and networking in will make you far more likely to get that job or you build a service that you can sell on that basis. And you think about all the aspects of selling from the onboarding to the customer sales, the SEO and others. You can use get lindy and things like that to have these agents. You can use N8N which allows you to build entire workflows of these AIs working in different ways or like I said you can just build things with your lovables, your II agents, your replets of the world for a few bucks with Google's AI studio. It's free. Like you have no excuse. You can build entire apps for free.
>> Yeah.
>> And again, once you show that off and you show off the ways that you use the AI, you will stand out from all the other candidates.
>> Totally. But most people listening to this who in that case won't do it because it's hard doing that first step.
>> Totally.
>> You know, but once you've got it, then you compound and you compound your capability.
>> Totally. I um I get the sense you're super ambitious, but would love to also understand what is actually a typical day look for you? So after I left Stability, I looked at every single meeting I had done in my entire time at Stability [laughter] and I went through every single one. I was like, does this add energy? Did it add anything to what I was trying to achieve or not? 99% of meetings didn't. So I do two meetings a week now regular uh like internal town hall and then a meeting with the team leaders. We've got a team of about 40 now. And then the rest of my time is just energy enhancing stuff. So I come and do some podcasts. I talk to interesting people and I just design and output stuff that then the teams take and we worked with them for years from stability and before and just implement like this is the most effective thing. Too many people take negative energy meetings or they don't focus on what they're good at. What I'm good at is building AI and systems and putting stuff together. So that's what I should spend the majority of my time doing.
>> Yeah. And who knows, you know, we we've mentioned uh on the show before and in this episode that are we going to all have our own mini agents who join these meetings for us in the next thousand days, probably next year, 2027. They handle all the calls for us. They're going to book dentist appointments. Are we is this going to be within the thousand days?
>> Yeah. [snorts] So again, there'll be the capability. Whether or not they're implemented is another question. But if you can extend yourself, even better. like I don't have a personal assistant anymore cuz I'm like I can book my dentist and things like that. I we want to reduce the number of layers between me and action effectively and I want to ensure that when I pass down instructions the team has the capability, agency and responsibility to get it done. So I think you know you need to find what your target is and then again see if your actions and other things are enhancing towards that and then leverage the technology to do it. So all my emails are read by an AI and then it surfaces the one I built custom and it surfaces the ones I should reply to instantly, you know, and the ones I can take a bit more time on. But then I batch process my emails. Like I just sit down, I just get through them and I do the next batch. If I need to go and talk to investors and things, I say no until I'm ready to go and talk to them. Like we haven't raised a big round yet cuz we didn't need to. When I do it, I'm going to do a very efficient process. I'm going to pre-prep them. I'll give them all the materials organized in the right way, customized to them using AI and their own investment thesis and then I'll go from there. So again, it's just rethinking what could I do if I had all these people, all these teams. Similarly with governments now with the Sage and we'll bring that to cities and other things as well. Rather than building a presentation, we're building a system that will personalize it to all their needs. And then you give them that, plus an agent that can talk them through it, plus interactive demo showing what does robotics and what does generative media do to Los Angeles and what do you need to know about it? And here are all the various ways you can do it. And it's in English and it's in Spanish and other things. And that will be far more impactful and likely to get them engaged than again the static approach, right? Going back to what we said at the beginning, it's only been a thousand days since all of this AI stuff.
has basically taken off. Um, if you weren't building what you're building now, or if you weren't building stability, what company would you be building? Would it be outside of AI?
>> I'd probably be building a fully vertically integrated robotics company.
>> Interesting. Like I think that it it's such a big breakthrough when you have embodied AI and you know the future of Westworld of you know I robot all of these things like if you look at the new unitary robot it's literally looks like the robot from my robot with Will Smith right these things will be among us and I don't see anyone building this with ethics morals other stuff like you look at the trolley problem you know like a train is hurtling down it's going to hit seven people and then you've got the person on the other side and there's one person on the track. you pull a lever to hit the one person to do you do that like these types of questions and the ethics morality just aren't captured there and we need sovereign capability in Europe for robotics for other things again as infrastructure that's the one side the other side is that it'd be really cool to build giant fighting Gundams you know like that would be massively entertaining you've got wreck robotics and robotics fighting league I think that's going to be such a growth sector it's going to be super cool you know
>> there you go that's your next [laughter] Yeah, we'll get to it.
>> When you see all of these Whimos going around San Francisco, what's your what's your take?
>> It's amazing when you get into one of these for the first time, right? You're like, "Wow."
>> Yeah.
>> And after a couple of times, it's like hydonic adaptation. You're like, "Of course." You know?
>> Yeah. [laughter]
>> Like again, it's the future. If cars were as safe, if human drivers were as safe as Whimo, we'd save 40,000 lives a year and a trillion dollars. And it's only going to get safer. Like there's going to be a point where insurance premiums for human drivers are just going to get so high human drivers won't be on the road. And self-driving cars like go Whimos. You look at Tesla Optimus. My guess is by 2028. Again, adjusting for Elon timelines and all that kind of stuff. A Tesla Optimus will be able to get into any car or any truck and drive it safer than a human. By get in, I mean like literally open the door, get in. That's for driver.
>> $20,000. So that's a that's a dollar an hour.
>> Yeah, it's crazy. I uh I was also just in San Francisco and just took my first whim. Shout out to my dad who just turned 60 has been in actually classic car insurance all his life and he was like what? And this is the future. I live here in Bangkok where the driving is crazy. You're from initially Bangladesh, right? Again, driving there crazy. Imagine if you can have self-driving uh cars in all of these crazy cities around the world. It's going to happen surely.
>> Well, I mean, all the Teslas will are being turned on to self-driving right now. And again, you'll have different formats once it's fully self-driving. You don't need a lot of the stuff that's currently in the cars. And it'll be a strictly better experience. Like, I'll have a car take as soon as we have a self-driving car here in London, I will have it drop my daughter off to school and bring her back, you know? So demand will increase for these services but the cost will decrease because it goes down to the constituent parts. So self-driving cars, we're working on self-driving economies in a way, right? Like again AI will be able to do it better. Self-driving healthcare, self-driving education.
>> Yeah. It's only going to it's only going to happen soon. Well, you can see it. And again, people are like, "Oh, it's not doing it yet." And that's cuz they're missing the wood for the trees. They don't see that all of the building blocks required are nearly there or are solvable with more computation with more intelligence with more applied and when you look at the costs it's tiny compared to
>> for sure
>> what exists today.
>> Um, to finish things off I think it'll be use I think a lot of our audience will be thinking what the hell is going to happen to my job in the next 2 to 3 years. Um, so our audience don't feel disheartened or they don't feel freaked out. What's most freaking you out right now in AI or what are you most concerned about? Well, apart from AI killing us all, which is a whole different topic, [laughter] it is this very human thing of, you know, our identities are tied up with our jobs and other stuff. Back in my home village in Bangladesh, not much is going to change. We won't have Optimus robots doing the farming and hanging out and stuff like that. You know, it's an agrarian society here in London. A lot of things will change, but what doesn't change is your connection to other people, how you identify yourself. If you're identifying yourself that you're an accountant, the AI is going to be better than you in a few years. If you're identifying yourself as part of your community with your family and others, again, when you think about a Star Trek future, it's one of abundance, particularly if you can harness this technology. And I worry that if we don't give everyone their own AIS that are looking out for them, that align them, that guide them through this, more and more people will feel lost, particularly against a backdrop of what is the British dream or the American dream or any of these things. We've lost that like the system is basically coming to an end there. So we have to tell better stories of the future otherwise again more and more people will get lost. There will be more and more polarization, violence, etc. Actually, a lot of AI CEOs I know because we do talk every so often have stopped doing public events altogether cuz after Charlie Kirk and things like that, they're like there will be this backlash against AI coming next year.
>> It was like uh is this is uh there was event where Samman was speaking and uh some guy like tried to give him give him a letter or something and that's just basic now. But
>> yeah, that's subpoena on stage. I think by this time next year, you will see very few AI leaders give any type of public talk. Again, I know specifically a bunch of AI leaders will no longer give public talks or do under very heavy security because they know the backlash is coming
>> because again
>> they talk about this stuff, but how many of them have actually put forward proper solutions? Like if you look at the intelligent internet white paper, you look at the lost economy, we lay out things for universal AI access, universal basic income, new ways money can flow, how to fix GDP, etc. And it's been really hard building that. It's as hard as building a rocket that takes off and lands itself. And that's why we're like, we've got to focus on this. And it might not be the best idea, but at least it's coherent and cohesive. Whereas most AI leaders are focused on let's build this technology to make profits, but when OpenAI says 200 billion of revenue, a 100red billion of that is AI agents that replace humans in their workforce. And you're like, okay, that's great for your bottom line, but what does it do to society? And they're like, well, it's not my responsibility. Just like Sam said, we're bringing in adult content because it's not for us to tell adults what to do. And you're like, well, there is a moral aspect to that in that you don't need to enable adult content that's super engaging, super addictive for 800 million users, cuz that will have real societal impact. Just like we have to look at stable diffusion and say, can this be used for bad things, deep fakes, etc. And how can we minimize that effect? Because the technology is inevitable. It doesn't mean that you can't be principled in the way that you use it and the way you apply it. Which is why we won the recent court cases because we didn't use any torrented or illegal content in it. You know, because we implemented safeguards even though it was at the expense of spread. So, it's a really tough environment when you have society this is that level of human impact and it's hard to come up solutions when you have a step change.
>> So much is going to change. Um, let's come back in 12 months. Love to have you back on the show and let's see how far how far things have progressed. But thank you for your time. This has been really interesting and hopefully gives our audience, you know, some insights into what's going to happen in the next 12 months. And I think what you guys are doing is really interesting and we'll include your uh latest book, new book in the description below. Go and check it out if you're new audience and we can't wait to see you next time.
>> Thank you. And again, everyone give your feedback. The book is free or 99 cents on Amazon. So, we really want people's feedback because we're trying the best we can. Thank you for having me on the show.
>> Amazing.