Microsoft AI CEO: The AI Future You’re Not Ready For

Welcome back to the Next Wave podcast where we dive into the ideas shaping the future of technology. Today's guest is someone who's been at the forefront of AI for over 15 years. He co founded DeepMind, he built Inflection AI and now he's leading Microsoft's massive copilot shift. Today we're talking to Mustafa Suleyman. In this conversation you'll hear why the so called AI training wall is a myth. How hallucinations might actually be a feature that we want, and why the next software moat isn't headcount or credentials. We also explore the agentic era of AI systems that don't just chat, but plan, click and get things done for you. And wait until you hear the wild story about copilot vision predicting the future of a flight that Mustafa was supposed to be on. By the end of this episode, you'll walk away with a practical playbook. You'll know what you can hand off to an AI agent this week, as well as how to think about moats in a world where anyone can ship software in minutes. So if you're living in AI or building right now, this one's for you. Let's get into it.

So I'm going to go ahead and start with your story a little bit. I know You've been in AI now for over 15 years with DeepMind and inflection and now Microsoft. I'm curious, what was it about Microsoft that drew you to them?

You know, it is quite incredible to think that Microsoft is 50 years old. Yeah. Like significantly older than me. And you know, for most of that time has been one of the leading tech companies in the world and also adapted to every new tech revolution. So each new wave requires the company to pretty much change its culture, sometimes changes business models, learn new programming languages, learn new go to market motions and stuff like that. So I think I was always, always inspired by the adaptability of it. And that's kind of what we have to do now. We have to adapt to this co pilot agentic AI companion era, which is just a totally different mindset to what the company's used to.

Yeah, absolutely. Have your beliefs or views on AI changed over the last 15 years, since the beginning of DeepMind to today?

Yeah, I think when I originally got involved in AI back in 2010, we were very, very obsessed with the AGI framing artificial general intelligence. Right. And I can still see that happening, but I think I've always been a bit more focused on what happens day to day practically with real users, you know, real consumers. Of technology. Right. And how that makes people feel, you know, rather than like the super intelligence that solves all the grand scientific problems. I still think that's super cool, but I can also see it really living life alongside us, being an integral part of my day to day workflow. As significant as your smartphone or your tablet or having access to the Internet.

Right, right. So I'm curious about the whole narrative around AI training. There was a narrative not too long ago that the AI training was going to hit this wall and that we would sort of run out of things to train on. And then we saw the test time compute and now they're sort of moving that inference and the amount of compute going to that other end of it. Do you think there is a wall and do you think like the, that the answer to this stuff is just throwing more compute at it all?

Yeah. It's interesting because the other side of that equation was people also saying that there wasn't going to be enough training data. Remember just before everyone's like, well pre training is done and you know, we're going to hit this wall and I think the lesson is that whatever obstacle, you know, comes up over a six to 12 month period, there are going to be five different, you know, ways that people are trying to make the models more efficient, use compute in a different way, generate data synthetically, which is what we're all doing now. We're producing human quality data completely generated by one of these models for reinforcement learning from AI feedback, from human feedback.

And if you look at how big the models have got, yes, they've got super huge, but they've also got tiny. You can now train a GPT3 capable model at 100 times smaller in terms of the inference cost, the cost that it takes you to produce an answer at test time than it was when it was originally trained. And that was only three years ago. So you got these like it's getting cheaper, easier, smaller, and it's also getting bigger, more powerful using more data and more compute. And that's what you expect when you know everything is working. It's like struck an oil well and now everywhere you look it's just like, okay, we have another amazing capability that's coming out.

So do you think that like throwing more compute at it and just kind of the bigger data centers, the more GPUs we put at it, that's sort of how we get to the next level or is the next level going to come from something entirely different?

Yeah, I think that trend is definitely going to continue. There's going to be larger models, more compute, and we will definitely, you know, continue to get massive gains out of that. But there's also like an emerging craft now which is in giving feedback to the models from other models.

Right.

So you get larger or more capable models, models to really efficiently teach smaller models with very high quality data to perform a specific task really well. And I think that that trend is also going to continue. And that's where you get this kind of efficient frontier. Like it's very expensive to build the biggest and best model first and then it's like 10x cheaper to build. What was Frontier 12 months ago today is kind of amazing. So then you have this entire swarm of second phase or second wave model development capabilities which are coming very cool.

Sort of stacking, reinforcement learning on top of what we're doing is kind of what's providing this sort of next era of AI.

Yeah. Simultaneously with just training larger models, getting higher quality data, feedback from other AIs, the kind of RL paradigm, we'll still use feedback from humans as well. So I think the lesson is that all of the capabilities end up cumulatively being valuable and you sort of chip away at the problem by both exploiting what you know to be rewarding and exploring new techniques and new methods that then get piled onto the stack.

Right, right. So I want to talk about hallucinations for a second. So hallucinations, I guess, can kind of be considered both a feature and a bug, depending on how you're using AI. Right. So if, you know, if you need something factual, you kind of want to eliminate hallucinations. But if you want something creative or you want to sort of solve new problems that nobody's solved before, hallucinations are something you want. How do we get to a place? Or do you see a pathway to getting to a place where we can trust what's coming out of these models?

One of the things that I find really interesting in technology is that sometimes a new trend gets a brand name or a label and then it confuses everybody about what's happened. You know, in some ways it causes like the entire field to hallucinate about what they're actually seeing. It's a really weird word, hallucinate, because if you think about the weaknesses of relational databases that have basically built pretty much all the value of software for 40 or 50 years, its big weakness is that it can't adapt, it can't change, it doesn't have any fuzziness. You only get precisely what you put in. Right. You only get out precisely what you put in. Whereas this now kind of solves a bit of that problem because it has this fuzzy, adaptive sort of abstraction that it can transfer knowledge from one domain to another, that it can interpolate. So you have two knowledge points and it can join those knowledge points by finding the gap in the middle, which is what you see when it's generating a new image that's in the style of something, for example. So hallucination is just an unfortunate word and I don't really see that as a long term problem. I mean, if you look at like three years ago, the models were really difficult to steer. Right. Stylistically, factually. They couldn't draw on citations or reference documents that are reliable. And now like two or three years later, with more compute, they're extremely controllable. They're very adherent to the behavior policies that we set for them. They're very responsive to the stylistic training. So they don't actually produce a huge amount of bias. They still get some things wrong, but it's very, very different. And so I do think that that is one of the most promising signs of the moment we're in, is that they're getting easier to control and steer as they get bigger.

Right.

Which I think was counterintuitive to a lot of people because a lot of people two or three years ago, like, oh, these are always going to be misinformation machines, they're going to be super chaotic, they're going to get it all wrong, cetera, et cetera. And that's turning out not to be true.

Yeah. And I feel like more and more now they're sort of grounding themselves with citations. Right. So if they're giving you factual information, you can see the citation and actually go and look at the source and sort of double check things if you need to.

Right. So that's the definition of trust.

Right.

If it knows when it doesn't know, or you know when you go and like verify a claim that it makes you feel confident that's a sensible source, you'll start to trust it more and more. And I think that's, you know, also very, very encouraging because the more people that use it, the better it will get. And then, you know, again, more people will use it because it's better quality.

Right, Right.

Everyone everywhere is talking about AI agents right now. But here's the thing, most companies are going about it all wrong. This guide cuts through the hype and shows you what's actually working right now. HubSpot has gathered insights from top industry leaders who Are implementing AI agents the right way? You'll discover which agent setups actually deliver ROI and how businesses are automating their marketing, sales and operations without replacing their teams. Get it right now by scanning the code or clicking the link in the description. Now let's get back to the show.

Now I know you mentioned AGI is not really the area that you tend to focus on, but I'm curious, do you have a definition for AGI? Because I feel like that term, the definition's still fuzzy and I feel like everybody who responds might have a little different definition. So I'm curious what your definition.

That's another challeng, right? Like hallucinations or. I mean words are difficult when you're sort of trying to predict the future. Like we don't actually have good definitions. So you know, people talk about super intelligence or general intelligence. You know, I like the method that we. Or the definition that we use to DeepMind, which is the ability to perform well across a wide range of environments. So the emphasis was on generality and you know, high quality performance where the threshold is, you know, maybe it's human level performance in some tasks, maybe it's all tasks, maybe it's exceeding human level performance. You know, I think we'll have to just wait and see that. The definition that I proposed in my book was artificial capable intelligence. Because what we all really care about is not some abstract idea of what intelligence is, which is really hard to define.

Right?

It's almost as hard as consciousness, it's really unclear. Or soul or spirit. Right? Very hard to define. But you can define capabilities, you know, like how much power does it draw? How many tokens does it produce per second? Does it solve this specific challenge? Can it retrieve over documents in the web? Can it take actions in your browser? Like, can it understand your scene and identify objects? When you use copilot vision to interact with the environment, they're very measurable and specific capabilities. So my mindset is always like this month, this quarter, what are the specific things that it can do? And that I think is a kind of practical way of handling that question. Very cool.

So I know you're familiar with Yann Lecun, right? He mentioned that he doesn't believe that LLMs are going to get us to AGI. He thinks it's going to be some other type of AI model. Do you see any other types of AI models after LLMs? What do you think is next beyond large language models?

I don't. I think that we already have the kernel of very powerful systems because tool use is a meta capability. Right. You know, that's kind of what has made us unique as a species. We're able to like pick things up, move them around, and then use that to manipulate our environment, to augment our intelligence. Right. It makes us more smart and capable that we can make fire or that we can use a stone ax, or that we can, you know, create a gpu. Right. Invent electricity. So we now have this system that obviously is nowhere near human level smart, but it can use tools. And we have a lot of very powerful software tools around, and a lot of those are not yet hooked up. So I think that even if you sort of stopped model development for a couple years ago, a core foundation model development, somehow, magically, but just theoretically, then we've already got this kind of technological overhang where the models have this, this amazing ability to use tools, talk to other AIs, source factual knowledge, orchestrate things in sequence. Right. And so, yeah, I don't see it slowing down anytime soon.

Gotcha. I want to just touch on ASI really quickly. I was listening to an interview with Naval Ravikant and he said that he believes ASI artificial superintelligence is a fantasy. We'll never actually see it. I'm just curious what your take is.

I don't know. I'm kind of one of these people who, whenever I hear never or fantasy, I'm just like, okay, well, it's a bit like the yarn thing, like maybe saying that these models will never scale to an AGI. I just think it's really hard to say for sure either way. And I think that our intuitions are changing every sort of year as we see these massive wave of new advances coming out. So I definitely think that there are certain capabilities with respect to ASI that we need to be a little bit more careful of. For example, naturally, if you just give one of these systems full autonomy with no human in the loop or on the loop or in any sense overseeing it's pretty powerful system. So I think we should think about these things in terms of what are the specific capabilities we want to accelerate and engineer into the models? What are the ones where we're not sure, you know, and potentially could have very, very powerful consequences if we mishandle them and really think about it from that perspective.

Right, right. I also think it's one of those areas where we might see those goalposts continually move and nobody really knows how to define it. So it's, you know, it's, it's hard to go there. Honestly. But what I want to talk about next is sort of some of the concerns that pop up. You know, a lot of the things that I see like in the YouTube comments and things like that when I make videos about AI and you know, one of the things that I see probably most often is AI is going to take all of our jobs. Right? AI is learning to code, AI is doing art. AI can be your graphic designer. Should people be worried about AI taking their job?

I do think the nature of work is going to change fundamentally. The thing is, no question about that. You know, we just as 50 years ago, there was no PC and everything was done on paper and we still relied on snail mail. You know, I mean, we are going through a big shift which is going to affect how we work, what we do, where we live. You know, that is a profound shift. And I think your day to day job has changed just as your job, you know, as a, you know, YouTuber and a media star now didn't really exist five years ago. Right, right. Every 10 years ago didn't really exist. So I think we've got to get ready for that transformation.

Gotcha. All right, so I want to, I want to ask really quickly about like software companies. I'd like maybe some advice from you on software companies because there was this whole narrative that software companies don't have a moat because everybody can just sort of prompt a software into existence. Or maybe one of the big incumbent tech companies might just build it in as a feature. What advice would you have to somebody who's trying to build a company right now?

Good question. I mean it depends on the size of your company. But for smaller companies, I think this is like an electric time. It's pretty unbelievable. I mean, I was just playing with this new version of GitHub Copilot now that has this unbelievable prompting ability way in natural language, you just give an instruction or you just say what you want, what kind of app or what kind of game, what kind of program you want, and then obviously it produces the code. You can then read all the code and ask for specific chunks of it to change. So in the process you're actually learning about how code is written, where the bugs are. It's very, very kind of insightful. And that iterative back and forth is a great way to prototype and actually build production grade code and ship the website, for example. So the barrier to entry has never been lower, which means experimentation by everybody is about to completely explode. The what that means is that it's going to be a very competitive Time for the big companies, medium companies, small companies, everybody. Which I think is an amazing thing for the consumer because we're going to get a lot of magical products and experiences. But I think it is going to be a very competitive and explosive, you know, hard time to really succeed in because it's going to be difficult to create those long term moats because things are going to change so, so quickly. But I think what we've seen over and over with these things is that people figure out ways of creating tremendous value and that value always gets paid out, you know, and people make, you know, big returns and have great outcomes. So I think it's a really positive story.

Yeah, I totally agree. So last question. I'm going to make it sort of a two parter. What excites you most about what you can do with AI today? And what excites you most about what we'll be able to do with AI in the coming years?

You know, one of the things I really enjoy doing is just having a conversation with copilot on voice on a commute, on the way home, like at the end of the day, just chatting through a problem, asking a question, rabbit holing around. Something I love basically learning. I love interactive dialogue. That's kind of the way I learn. I mean I love reading as well. But in some ways my books and my ideas just come alive because I'm able to kind of, you know, run down any corner of a topic with copilot. So I do love that experience. I've also been using the copilot vision experience a lot and it sees everything that you see in real time. You can talk to it about exactly what you're both seeing. And that is quite a magical experience. I was in LAX the other day and I opened copilot vision and it knew which FI was in. It looked at the gate number. It was like, oh, it looks like you're heading to seattle because it's 4:30 on a Sunday. So there's only one flight right now. By the way, did you know that your flight is delayed? Oh wow. And I was like, it's not delayed, co pilot, you've obviously got it wrong. Hallucination. So then I went to the front desk and I was like, is the flight delayed? And they were like, oh yeah, we're just about to make an absolute.

So it knew before they did. Yeah, amazing.

Like that's a magical experience. It is crazy. It's incredible. So the things I'm really excited about coming up are the actions. Copilot actions is mind blowing. I mean, it will really just operate on your desktop in Windows, highlighting areas where you're stuck pointing at things, helping you with settings, or in your browser like, we'll buy things, book things, plan ahead. It's quite magical to see Copilot moving the mouse around, clicking on things, opening new tabs, entering stuff in forms. You're just like, we're living in the future.

Yeah, yeah, yeah.

Amazing.

Sounds like problem solving in Windows is going to get automated pretty quickly here. Exactly. Amazing. Well, thank you so much for spending the time with me. Has been a fascinating interview. I really, really appreciate you spending the time.

Pleasure, man. Thanks a lot.

It's really fun.

Cheers.

If you remember one thing from today, it's that agents that do beat apps that wait. That's the shift. So here's your challenge this week. Pick one repetitive task. It could be booking a meeting, filling out a form, summarizing notes, or cross checking data and hand it off to an AI agent. Then let us know in the comments how it went. All the links and resources are in the description and if this episode helped you, make sure to subscribe and share with a builder who needs this playbook. Thanks so much for tuning in and hopefully we'll see you in the next one.