Emad Mostaque - Money for the Intelligence Age - TOKEN2049 Dubai 2025

It's exciting time I think for everyone and I think we're at the cusp of an intelligence revolution. Um, this is a chart of AI last year and you know last year I was just leaving my former company I founded stability AI where we had 300 million downloads of our models best known for stable diffusion that kicked off the image generation trend but we were building models for video audio protein folding and more. And the IQ of the top AIS was 90 today it's 120. How many of you here have an IQ over 120? Now you're all too modest.

You know, the IIS now are as dumb as they'll ever be. And this 03 level AI is going to be available to everyone everywhere, even on devices within the next couple of years. This isn't something we're really ready for because when you think about the world and intelligence, this is a chart of click. Okay. Wrong button.

My fault. uh AI wouldn't make that mistake. This is a chart of intelligence and IQ across countries which is really a chart of energy and infrastructure. It isn't that people are dumber than some places than others. Just places like China for example rapidly ramped up the intelligence by building an infrastructure for education.

Now though everyone everywhere around the world will be able to have 120 IQ buddy that helps them and in fact a swarm of them. And what does that mean for the world? What does it mean for the economy? How does that look? This is OpenAI's latest 03 model on coding.

It's now in the top 100 coders in the world on the competitive code force rankings. It's a much better coder than me. It's a much better coder than you. This is what it looks like with Google's new AIM model. It's AI medical expert.

The interesting thing here is that they looked at clinicians, then clinicians assisted by AI, and then AI by itself. The AI by itself outperforms humans with AI. And this is a shocking revelation. AI can also work longer and harder. Previously, a year or two ago, when you talk to an AI, it was like talking to a stochastic parrot.

It didn't have any memory. It just bounced back the ideas to you. Now, with the new thinking models, the R1's and more, AI can think longer and longer. It's more like a teammate that you can set off and say, "Do deep research on this topic, see what it looks like, and come back to me with a vision board. Go and do some research here."

And that will just increase. It's doubling every seven months. So, within a year or two, you'll be interacting with your eyes on Zoom. They'll be WhatsAppapping it and saying, "Go and do work for me. They will be your teammates because the hardest thing to scale is the team."

But all of a sudden you will have these amazing cooks that just get the job done and never let you down. What does that mean for the economy and society? You know, uh this is from the IMF. So the three lines are traditional baseline AGI and aggressive AGI. Now the great thing is that output goes up as we can build better teams.

The concern is well look at wages. It's not very happy. How will you compete with artificial intelligences? How will your children compete with artificial intelligences? AI is going to build new economies that are more efficient than the human economy and they will be everywhere.

And this is the final year before they become ubiquitous. It's crazy to think about because what we've classically had is the ability to scale our organizations has been dependent on labor and then labor gets capital in exchange. But what happens when things change? This is agricultural versus non-aggricultural employment. There aren't that many farmers now compared to how they used to be.

Now, what happens when you can build teams of AIs that are more efficient than anyone? The labor capital connection is broken. For a company to scale, an organization to scale, you just need more capital, more compute. And then what does that mean for money? When a country's capacity was previously industrial stock, its talent, it's teaching.

Now, it's how many GPUs do you have? How much intelligence do you have? And how is it customized to you? Because a generalized intelligence is one thing. We have this concept called AI Atlantis.

We discover this new continent of a trillion AI graduates, a billion robots, and they're coming. But when you have a graduate, when you customize it to your own needs and yourself as a member of your team, then it's the most impactful. When it has a memory and it's not forgetful or sarcastic parrot, that's when it's useful. And you need that because these AIs are more persuasive than anything. Now, in fact, you know, we've had this thing of self-s sovereign money previously, value transfer rails, and Bitcoin was the original.

We need to now move to self-s sovereign intelligence because some of you may have seen this weekend, OpenAI became sick of antic, and it started glazing everyone. And that was weird. You know, some of you may have also seen a Reddit study just came out. They let agents loose on Reddit and they created all sorts of personas and that was weird in itself because it manipulated people but a claw-based model was a 98th percentile in convincingness persuasion. The models have massive biases in them.

Dan Hendris who's advised to scale AI and XAI did a study where he said you know the trolley problem where you pull a trolley it's heading towards one person or five people you decide the direction. He put that to the top AIs and what did the AI say for an American life versus a Nigerian life? 10 American lives to one Nigerian life because OpenAI paid all the Nigerians to label the data and the inherent biases came out. What happens when you have AI in the government? The Trump tariffs were calculated with chat GPT.

That's a bit weird and it's only going to get weirder. So now we have AI entering every part of our lives surrounding us. What's inside these models? How does it work? How does the labor capital divide adjust?

I'm not quite sure to be honest. I say money and intelligence age, but I've been working towards that because I think it's important. And I think what we ultimately want for the AIs that matter, the AIs that help us through our days, that educate our kid, that manage our medical system, that run our governments. Like, would anyone here think that AI in a few years will do a worse job than politicians? Well, no.

Zero hands up. or you might just be lazy. You know, definitely this will be in every part of our lives. Checking every medical decision first and making every medical decision. We need the elements of blockchain for that verification, coordination, and resilience.

You don't want it to go down one day or to be changed up. Meta and Google are selling ad space within their open language models now. So, you want a beer, it's a Bud Light. Think about that again. It's every part of your society.

As your children's best friend is an AI, we need something better. Our solution as part of the solution is the intelligent internet. We think that the AI that does these important things that impact lives need to have open data and open models. They can't be black boxes. Anthropic did a study called sleeper agents where they showed with a few thousand words and trillions of words of input you could train a model so that with a code word it turned evil on demand.

You can't have that vulnerability around you. And so you need to have a distributed system. We need universal AI for the people UI. We don't need UBI necessarily. Having access to 120 IQ model then makes about your own agency, your own sovereignty.

We see people all around the world now using raplet and other systems to build things. Doesn't matter where they were. In fact, at stability, we had 80 top developers building state-of-the-art models, which is hard. A third of them only had a few years of AI experience and they threw themselves in. AI is actually much easier than crypto to get into and it's a lot less mentally damaging.

You know, it's fun. Just code, vibe code. In fact, everyone here should try replet or one of these systems. So, the system that we decided upon was that there needs to be one that built supercomputers for our common knowledge of humanity. What do you need to know for cancer?

And who's organizing the cancer knowledge in the world? What does the optimal education stack look like? What's the best way to govern a country or to run financial rails? We're going to spin up supercomputers for all of those. On the national side though, you need to have AI owned by the people, for the people, for the UAE's education system, for Malaysia's education system needs to be customized to those local.

And the supercompute for healthcare in any of these countries, education, government is only going to go up. So let's enable that. So you have the creation of knowledge at the top and the coordination and the customization. The final point is edge. All of you need your own AIS that you own yourselves.

run on your smartphone, run on your local computers, run in your communities that represent you and are aligned with you. They don't care about someone else's business. And again, I think this is important for all the regulated industries. They need to be open- source open data. So, we designed a system with all of these.

And the first thing was how do we build better models? How to build better data sets. So, a few weeks ago, we released II thought, which is our reasoning data set. This is like DeepSeek. So, DeepSeek released an R1 model that tuned base models up.

I thought outperforms it on all metrics and it's freely available everything as MIT. So, we figured out how to make data better. In fact, it uses half the number of tokens for thinking for the same output and that's the importance of quality data and we want quality data in these models for medicine, from education, for government and more. We just released uh open deep researcher again fully open source feel free to download it use it as you like. This is OpenAI's deep research complexity deep research on steroids.

So we use a lot of the things that eventually made their way into 03 but available for any model wrapping and you can see 72% accuracy versus 67% 52%. So feel free to use that. Today we're releasing the world's best open edge medical model. So this runs on device and it's almost at doctor level. I don't think it's suitable for medical use yet and it will be better.

But I can say that within a year on your edge device, you will have a more empathetic AI that's better than any doctor and it'll be freely available to everyone. But then that'll be wrapped in an agent and a researcher. And we've got some more announcements to come that will help everyone who's going through their cancer journey through their cancer journey because we can't scale empathy. It will help teach every kid on an individualized basis because are you a visual thinker? Are you auditory thinker?

Maybe we'll have an AI that checks the stupid laws so they're not so stupid. Common sense AI might do well. And I think this is infrastructure for society. I think that as you move to a post labor economy, this needs to be accessible and it's only about your own ability. People don't want handouts.

They want the ability to have a level playing field and finally you can get the expertise you need to scale. We've seen companies like Midjourney, like Curser and others scale to hundreds of millions of revenue with just a few people. The things that you all imagine that you could build to make a real difference, you can do now. I just say on a panel and you'll know they're not North Korean hackers who your dev team now. I think this is one of the most exciting and scary times ever.

But we've got to build these models and these systems. We've had this thing in crypto, not your keys, not your crypto. I think the next thing is not your models, not your mind. The models that matter and you'll have your Apple intelligence and your super intelligence as well. Those are all there, but we need to make sure these are verifiable, coordinated, and resilient.

The stuff that runs our society, the infrastructure, and we need to make sure it's available as widely as possible and localized as possible because that's what our kids deserve. That's what our families deserve. And I think that now is the ideal time to make sure this is decentralized and open as opposed to controlled by a few who can decide how your brain works. So I didn't give the answer of money in an intelligence age. Unfortunately 15 minutes isn't enough for that, but we do have our white paper and a bunch of research coming out soon.

So I encourage you all to sign up for ii.inc. The build is here. Please go and take our technology and build. We'll re everything MIT.

Let us know how we can improve. We want to build the technology for humanity. We're going to spin up teams in every country to build the localized stacks. And we hope we can incentivize this so we can have AI for the people, by the people, and of the people. Thank you everyone and take care.