Mustafa Suleyman on Microsoft’s Humanist Superintelligence Bet

0:00We're here at Build 2026 in San Francisco and you had some big announcements today. Seven new models in one day, eh? We did, yeah. No, we're very, very excited. It's been a very intense six months. We actually have to be truly self-sufficient in AI. We're able to train our own models at the absolute frontier. These are all the basic building blocks of a superintelligence. It needs to be able to hear. It needs to be able to talk. It needs to be able to see. have a VLM. It needs to be able to reason or think. It obviously has to be able to write world-class code. But the motivation is, does this accelerate human progress?

0:36That's the test. And if it doesn't do that, then technology should be rejected. Every clinician, whether nurse or doctor or physiotherapist, is going to have an AI-assisted tool that is helping them with their diagnosis, with their workflow, with their decision-making 24-7. Welcome humans to the Neuron AI Explained. I'm your host, Corey Knowles, and today I'm here with a guest we've had before, Mustafa Suleiman, CEO of Microsoft AI. How are you today, Mustafa? Doing great, man. Thank you. Thanks for doing this again. Absolutely. Excited to have you because we're here at Build 2026 in San Francisco, and you had some big announcements today.

1:13Seven new models in one day, eh? We did, yeah. No, we're very, very excited. It's been a very intense six months, and we now have seven new models. Our transcription model is the best in the world, state of the art it's also the fastest most cost effective of any hyperscaler inside of azure which is very exciting our voice generation model is out now voice generation 2 super awesome model we also have our image to image editing image to image model and our image to image editing model wow two additional image models they're now number two and number three on the leaderboards being out nano banana 2 and all of the other models so we're very very proud of that and we also have our code flash model and our thinking model.

1:57So it's been quite a journey, but we're very excited. That's a really big step up. You guys had a lot of restructuring around the AI team and things back into the year. And I feel like, is that a thing that has kind of led into the ability to make today happen? Yeah, totally. I mean, basically for much of last year, we were sort of negotiating our contracts for the next phase with OpenAI, and we're very happy with how it landed. We're partnering with them for many years to come, but we also set a drop dead date, which meant that we actually have to be truly self-sufficient in AI, be able to train our own models of the absolute frontier.

2:35And so since October, we were sort of freed up to pursue true superintelligence internally. And that's meant that I could deploy all my crew on building the best models in the world. And that's what we've been up to. That's really awesome. Let's talk a little about MAI Thinking 1. So this is your first foray into reasoning models. How did that come about? How long have y 'all been working on this? Yeah. So, I mean, the way to think about it is that, you know, the last few years, everyone's been sort of focused on pre-trained models, and that's very, very important. And we know that our enterprises and our developers really cared about coding.

3:10Yeah. And some of the other models that we get from third parties, they're optimizing for consumer use cases. So they have like the long tail of languages around the world. They have cultural entertainment information and so on. And that chews up tokens. And so instead, our pre-trained model is 50 % code, really high quality, commercial grade, licensed in exactly the right way, de-duplicated, decontaminated, really focused on security. And that has produced a very good reasoning model, right? So that's actually a very, very important thing. It's very distinct to, say, GDF, Google, or OpenAI. Yeah.

3:51And we did it deliberately so that the model could be great at maths and science and the kind of STEM subjects. Yeah. Because once it learns how to reason, it kind of has this abstract general idea of the logical relation of things, and then it can use that to learn in new domains. And that produces a general purpose thinking agent that any developer can then adapt to a downstream agentic task. Where you want it to have some knowledge of mathematics and of physics, for example, to help around its view it's bringing into other things, I guess. Yeah, that's actually what drives it to do really well in RL.

4:27Because much of what happens in RL when you climb a model reinforcement learning style in a broad, open environment is that the model has to make decisions. And so it needs logic to be able to make those decisions. Should it, you know, try this tool or that tool? Should it, you know, invoke an API? Should it generate novel code? And it needs to obviously do that over many, many time steps accurately. And that's general purpose reasoning, which is what MAI is thinking one is exceptionally good at now. That's really awesome. That's really awesome. And you said you see this as a, I think you called it a medium weight?

5:00Medium weight class. Yeah, it's a 35 billion active parameter, 1 trillion tokens, 256K context length. We've actually published an extremely comprehensive technical report today. 109 pages in great detail going through not just the algorithms, but the data mix. Yeah. All the training methods that we use. We've actually shared a lot of lessons that we've picked up along the way, things that we haven't got right that we've had to improve. You know, so it's a really comprehensive report that shares everything that we've been working on. That's really awesome. And I think that people sleep on in talking about these.

5:38when you talk about a million parameters, a trillion parameters, is that data doesn't just create itself. That's a massive undertaking to come up with that amount of data. Are there teams working on that alone? We actually have more than 250 trillion tokens internally. Wow. And we whittle that down, at least for MAI Thinking 1, to 30 trillion tokens. And so we're constantly running ablations. Ablations are like pairwise comparisons between this set of data and that set of data. And we do that at huge scale to find the optimal combination of tokens to fit inside the medium-sized weight window. Because obviously, we want these things to be affordable to run inference on.

6:19They can't just be giant because that's not the goal. The goal is to make them hyper-inference efficient whilst delivering the right performance that we need. That makes sense. And it's really fascinating that I think you've hit at an interesting time here where I've, you know, and I mentioned this to you before we talk, in our early year predictions, one of the things I thought was that I thought Microsoft seems uniquely keyed up to have a number of things coming together because it's felt like we were seeing pieces of that come out with, you know, there's an image model and then there's a speech model and then there's a voice model and it's like, kind of like watching something be built in public, if that makes sense.

6:53I mean, these are all the basic building blocks of a superintelligence. Yes. It needs to be able to hear. It needs to be able to talk. It needs to be able to see, have a VLM. It needs to be able to reason or think. It obviously has to be able to write world-class code. And there's a few other very important capabilities which are coming down the track in the next three or four months. And then they need to all be brought together in very efficient, agentic forms that allow them to go operate in the real world. And I think that's kind of what we're working towards in the next few months. That's really awesome.

7:26I'm anxious to see that and see how that comes out. And now a message from this video's sponsor, BeyondTrust. Identity is one of the biggest attack surfaces in your business, and it's getting messier fast. You've got employees, cloud apps, SaaS tools, machine identities, and now AI agents all connecting into the same environment. That means risky privileges and hidden attack paths can sit quietly in the background until someone finds them the hard way. Beyond Trust's free identity security risk assessment gives you a clear snapshot of those risks in just 24 hours. It helps uncover excessive privileges, toxic access, and identity security gaps across cloud, SaaS, and hybrid environments.

8:07To start your free Beyond Trust's identity security risk assessment today, just click the link in the description below. And now, back to the episode. in thinking of superintelligence and this quest for something bigger. You talked this morning about humanistic superintelligence, and I was wondering if you could talk to us a little about that idea, what it means, and how it came together. It's a very important design intent. I've always been inspired by science. Science and technology is really the engine of everything that we are as humans. It's true. All of our progress. and that's the real quest here you know we we want to invent the next big breakthrough that makes us all healthier and happier and live better lives not just for us but for seven you know eight billion people across the globe and so my opinion the way to do that is to train these models to be really good generalist reasoners that we can use to invent new science and that's i think got to be the motivation of building these models.

9:10The motivation is, does this accelerate human progress? That's the test. And if it doesn't do that, then technology should be rejected. Technology for its own sake isn't an inherent good. Technology applied in the right ways to make us humans better and healthier. That's really the kind of benchmark. That's the goal. And so the purpose of framing this in humanist terms is to remind everybody continually that that's the project. That's the project. And we have to put technology to that test because people are anxious at the moment, understandably. There's a lot of uncertainty and we don't quite know how it's going to play out.

9:48I think we have to be upfront and straightforward about that. And we have to be very public about what we're trying to do. I mean, not everybody's trying to do that. Not everybody declares it in those terms. And we still have a lot of work to do to make good on that promise, but we're heading in the right direction. Agreed. And I tend to feel like a little ways farther out, I think a good outcome is pretty certain that there's this window of time that's very, very in flux. And, you know, I mean, none of us have a crystal ball, but it's good to see that you're making decisions that are centered around that idea.

10:20And I think that's very important. Yeah. And I mean, we get to choose what applications we work on, right? So today we also announced that we have this incredible partnership with Mayo Clinic. Yes. It's a really big deal. I mean, number one hospital system in the world. Many people associate them with kind of elite care for the people who can afford it. Yeah. Truth is, 65 % of their patients are on Medicare, Medicaid. And that's actually remarkable. So they have a very representative population, cohort in their patient population. They have the largest, most digitized longitudinal patient record of anyone in the world, including any government.

11:00So it's the highest quality data that exists. And, you know, I think that's going to really make a huge difference. It's multimodal. It has genomics. It has clinical practice as well. And so we're really, really excited about this because I think that it's going to really help with diagnostics, with treatment, with preventative intervention. And it's going to take us quite a few years to really land the model here, but we're going to train a new foundation model from scratch together. So it's very exciting. That's amazing. And, you know, the thing that's really cool about it is it also creates access to male caliber information.

11:36Like they've, you know, they've done so much research work with rare conditions and diseases and things that don't get tons and tons of attention, as well as, you know, the many cancers and other diseases we hear about. So, I mean, I can't imagine that the data they have is not a goldmine of stuff for longevity, for quality of life improvements. Yeah. And they've already deployed hundreds of AI diagnostic algorithms themselves. So, I mean, this is kind of an important thing to realize. They invested in digital pathology seven or eight years ago. They've been pushing mammography, radiology, you know, from day one.

12:15They've been doing digitized radiotherapy. Like they've really been pushing the boundaries for a very long time. So they're a great partner for us because we're already starting at the absolute peak at the top. And I think that's the big quest. It's like, you know, obviously we want to make Mayo quality health care available to everybody. Yeah. But we also want to, you know, uncover new diagnosis, new correlations that were maybe surprising because we have this like breadth of very longitudinal multimodal data. Just by existing, ideally, there will be things that it can uncover that thinking of AI in terms of pattern recognition and things like that.

12:52There's so much something like that could uncover just in working with itself. Yeah, totally. I mean, just as every coder today now has an AI-assisted coding agent, every clinician, whether nurse or doctor or physiotherapist, is going to have an AI-assisted tool that is helping them with their diagnosis, with their workflow, with their decision-making 24-7. Yeah. So I'm pretty sure that's going to be the next big product market fit explosion in the application of AI, just as we've seen chat and code over the last three or four years. Yeah. Healthcare, I think in 2027, you know, is probably going to be the next one.

13:30I think you're right. And can't come soon enough to suit me. The truth is, that's, you know, when you think of the promise of AI, that is what brought so many people into this space. It's not just this idea of building cool software and building this this thing necessarily. But like at the core of all of that is this idea that this can help with the problems humans can't solve on their own, like problems around longevity, around not just longevity, but making life something you want to continue living, you know, making it good enough and finding it. finding new ways to use exciting things. And I think it's excellent that you guys are doing this work, and I'm excited to see where it goes.

14:06Thank you. Yeah, me too. Very exciting. Mustafa, thank you so much for joining us again. It's been an absolute pleasure. I look forward to seeing what you all do, and can't wait to get my hands on some of the new toys you dropped today. Pleasure, man. Thanks very much. I appreciate it. Thank you. If you haven't yet, please take just a moment to like and subscribe. We'd love to have you, and pop by the neuron.ai, sign up for our email newsletter that goes out to 700 ,000 or so every morning. We'd love to count you as one of our readers. And on that note, that's it for today. Farewell for now, humans.