Is AI a new species? Microsoft’s Mustafa Suleyman thinks so

0:00Scaling a business in the US can feel like a maze. Each state has its own payroll, benefits, and compliance rules, pulling your focus away from growth, which is why founders use Deal. Deal is the professional employer organization, or PEO, that gives you a dedicated HR expert, plus Fortune 500-level benefits for your team. The National Association of PEOs says businesses can grow twice as fast if they use one. So, if you're scaling, make it simple with Deal. Go to deel.com. slash MOS and get up to three months free. If you're in private capital, you understand that your next great deal starts with who you know.

0:41That's why over 3 ,000 firms trust Affinity, the CRM platform purpose-built for private capital. Affinity's relationship intelligence reveals your team's strongest paths to founders, investors, and decision makers using insights It's hidden in emails, calendars, and meetings. With powerful AI tools like Affinity NoteTaker and Deal Assist, you'll spend less time managing data and more time closing the right deals. Learn more at affinity.co slash marketplace. Hi, listeners. This week, we're sharing my conversation with Mustafa Suleiman from the Masters of Scale Summit. Mustafa is the CEO of Microsoft AI, and it was inspiring and thought-provoking to talk with him about the transformational moment we're in.

1:33I hope you enjoy this chat we had live on stage at the Presidio Theater in San Francisco.

1:52You've got to have incredible talent at every position. It's like this huge push. There are fires burning when you're going home. Can you believe it? Such an idiot. And then you go back to, this is totally going to be amazing. There are so many easy ways. I have no idea what to do. Sorry, we made a mistake. but you have to time it right. Oops. Working out of a three-bedroom apartment. Stuff that just seems absolutely nut balls. Ten years later, we're like, well, that's just how you do it. We haven't made it just how you do it.

2:22This is Masters of Scale.

2:32So you, other people, have sometimes, you know, made parallels between AI and a species. and thinking about this as starting small, like I said. Let's go. Yes. How is that as a lens for thinking about AI? What are the places in which it's a good lens? What are the places in which it's a misleading lens? What's the way that it should guide kind of the global thinking for how we're getting there? You know, I think when we have something that is fundamentally new, that is unlike anything we've seen before, which each new wave of technology really does feel like that. Think about how utterly magical and crazy it must have been to have electricity for the first time, or even to speak to somebody via a telephone line across the Atlantic.

3:24It must have just been mind-blowing. It adds a completely new mental representation to your worldview of what's possible. So each time that happens, we sort of struggle for the right metaphor to relate it to something that we do know. It isn't ultimately going to be like that thing that we do know, but it's the best that we have before it arrives. And I propose this new digital species metaphor just because when you step back and look at the capabilities of these things, it's really the closest corollary, even though that kind of raises a lot of things that we don't want it to be and I think frames the question of containment in the correct way.

4:06I mean, these models will be able to see what you see, hear what you hear, understand and interact with text in real time, take actions on your behalf. Those capabilities are now coming into vogue. And I think the right metaphor, the most similar alternative that we've got is the species. And that comes with, I think, a helpful frame for thinking about what we don't want it to be as well. And what would you say would be on the species, the one thing that it's really important that we do and the one thing that's really important that we try not to do in order to guide into digital species as we're...

4:44And by the way, I recommend Mustafa's book, The Coming Wave, which also goes in as depth. This is like the 60-second version. I think one of the things that's incredible about these models is that they don't give you exactly what you put in. I mean, that's sort of the great ambition of software, right? We want it to tell me something that I don't know. And so hallucinations, I think, are a kind of unfortunate phrase. Like, that's not a downside. To me, that's the upside. Called creativity. Yeah, creativity. We want like a wide variety of possible responses given some input. And that malleability and that ambiguity is what we want.

5:26And so having them learn their own representations of things rather than us handcraft those features, which was the core motivation of the last 15 years of machine learning, it's great that it's now doing that. But what we sort of need to figure out is where is the boundary on that learning, right? At the moment, there's very little, if any, recursive self-improvement. So there's not a closed loop of self-improvement that doesn't have a human directly overseeing it. But we can see that coming into view in 2025. Teams will start experimenting with that. And so I think that's one to watch as something to be cautious about.

6:01The other thing is just straight up autonomy, right? Like, you know, it's clearly going to increase the risk if these models can, you know, interact in arbitrary digital environments, spin up their own VMs, you know, take actions on, you know, web pages, interact with APIs, and do all that kind of thing completely independently of human oversight and control. So those are two capabilities that we would be pretty concerned about. And then on the positive side? I think that like the positive side is that they will be immensely creative. And I think that they're going to help us to interact with the very best of ourselves.

6:39You know, the fact that they don't need to be, if well-designed, snarky, judgmental, shame-inducing. You know, most humans can be assholes, right? There's no reason for these things to be mean. Some people will program some AI companions to be all of those things. And that isn't an inevitable outcome. That is a choice that some designers will make. And, you know, I think structurally we should do everything in our power to limit those kinds of things in the ecosystem, in the norms and values and stuff. But some people will do it. But I think that there's plenty of space for these things to really help us be our best selves.

7:19Like I read this paper three weeks ago that basically reported that a bunch of people who had real conspiracy theories. I mean, I'm talking about like flat earth level conspiracy theory. There's a few of those. Yeah. I guess stay away from election stuff but just even like i think that we should really all agree that the flat earth one is like that's pretty nuts yes so like people who had chatted to a chatbot over the an extended period of time like i think it was like six weeks or something reduced their propensity to believe the conspiracy theory and that's because the chatbot is patient it's non-judgmental yes it doesn't put you down it's relentless yes it always comes back and um you know generally draws on you know the kind of scientific literature with an evidence based on and so forth so i think there are very promising signs that the upside really will be incredible so um actually i'm going to jump ahead to a question that i was planning on asking you in a bit but i think this is a good frame for it which is you know when you karen and i started inflection one of the founding principles was that eq um as as important as iq right say a little bit about why what that meant for pi you know what what is the thinking of that and why is that important across the board not just for what pi is doing right right um i mean you know iq i think we can generally consider to be the sort of accuracy um the speed the comprehensiveness the relevancy of an answer the extent to which it has real-time access to information you know all of those things we're sort of making there's a steady march of progress and i think what i sort of noticed was that what people tended to, you know, people in the community of sort of AI researchers in general tended to neglect the importance of the delivery vehicle for the information.

9:09Like, it's kind of a very nerdy thing to just say, well, if I just lay out the facts, then people will clearly see that this is correct. Yes, the engineering mindset. Yes. And it turns out that actually tone, style, you know, the kind of emotional intelligence of these models, the extent to which they will ask you questions, you know, the extent to which it reflects back in a sort of type of language that you might use and so on. That delivery vehicle for the substance is perhaps more important to the majority of consumers than just an objective regurgitation of Wikipedia. And so I think that's going to be one of the key capabilities that I think everyone's starting to wrestle with that now is like this kind of agentic future isn't just about the actions.

9:52People can clearly see it's also about the personality. And I'm very interested in how we sort of engineer personality, because I think that's kind of what people are really going to find, what I can see people find very valuable. So speaking of the agentic future, give us a bit of a lens into it. Like, what is the places and how you're thinking about it from a co-pilot perspective? What is the kind of ways that you think about are the likelihood in the next two to five years about how agents will be playing a role in our life? What's important, you know, from the species level now down to the specific about these agents and how should we navigate with them?

10:32So the first step for the agentic future is that your co-pilot has to, your AI companion in general has to see what you see. And having an aide or an assistant or a companion that is really seeing the pixels that you see on the screen in your browser on your desktop on your phone means that there's a kind of level of constant you know sort of awareness of your sensory input that enables your companion to also observe what you're seeing and then you can say you can you can sort of use ambiguous references like like remember that thing that I saw? Or where were those things? And that is a kind of level of sort of understanding that we've sort of never had before.

11:16And it enables your sort of AI to then, you know, sort of act on your behalf, right? And that means like, navigating on the browser, it means using APIs, booking things, buying and planning. And I think that, you know, that we've obviously got a lot of cool demos floating around of those kinds of things at the moment. It seems to me that we're still a little way away of getting those ready for in production. You can see it from all the previous waves, like just before GPT-3, there were models, LLMs inside the big companies and stuff. And that was probably in sort of 2020, 2021. And they were really flaky.

11:52And I think that may be where we are with the AQ side of things, the actions quotient. So I think that getting things to work 50, 60 % of the time is, you know, great. And we have to get them to like 99 % accuracy. And you can sort of see that with voice recognition and dictation. Like that's been a 15 or 20 year trajectory. And it's really only in the last sort of two, three, four years where it's crossed the threshold where it's sort of like 99.5-ish percent accurate. It's personalized, you know. And so, you know, you're starting to see a huge uptick in people going voice first, partly because of the input, but also because of the generation.

12:29So I think it's a few years away from that yet. And what do you think is the intersection between the voice input? Because I actually completely agree. I actually think part of, and you know this because we've talked about this a lot, but it's the generative AI revolution allows it to be in that conversation that allows the voice input to work so better because you can just speak to it and then it can interpret actually what you're saying. And how is that going to kind of bring that extra elevation to the agentic universe? Yeah. The interface, the shape of the interface, just very abstractly, governs what you can put into it.

13:07So because the search box in a search engine was just a letterbox, we learned to speak the language of search, right? We compressed our ideas into like a three or four or five word. It's not even a sentence. 1.6 on average, just to be clear. There you go. Even shorter. Yeah, maybe 1.6. so and I think what's interesting about these voice experiences is that it sort of unlocks a new part of your mind when interacting with a computer because you can speak in full sentences because you can sort of self-correct you can sort of go forward and back and you know you can add in all these the other junk that we have when we're just kind of you know talking off the top of our head and then the model speaks back to you in like paragraphs it you suddenly think to ask and talk about things which you never would have digitized previously.

13:55So that I think is probably a good, I mean, I think it's pretty sure to be a good framework that tells you what is likely to happen on the action side of things. Because you have this always available AI companion that can really do any digital task you can do, I think you will ask it to do things that you don't do yourself on the computer today. And that I think is a big shift because the barrier to entry to get something done is about to go through the floor, right? It's both because it's zero marginal cost and because of the friction has like really diminished. And so then you'll think of things that you hadn't thought about to do yourself because it's just too much of a pain.

14:35More with Mustafa Suleiman in just a minute.

14:43Real leaders don't back down when the stakes are high. They innovate, they push forward, and then they take the stage at the Masters of Scale Summit. Join us in San Francisco, October 7th to 9th, to hear from the CEO of the New York Times, scientists using cutting-edge technology to find cures, the leader of crypto powerhouse Coinbase, a retired four-star general, and many, many more. Apply now at mastersofscale.com slash apply25. That's mastersofscale.com slash apply25. If you're ready to take your startup from idea to impact, then AWS is your launchpad. From data storage to machine learning to secure app hosting, AWS gives you the tech trusted by the world's fastest scaling startups.

15:32And here's the best part. Join AWS Activate today and you could score up to$100 ,000 in AWS credits tailored to your stage and network. Go to aws.amazon.com slash activate and start building. Expanding your business in the U.S. can feel like a maze. Every state has its own payroll, benefits, and compliance rules, which can pull your focus away from growth. That's why founders use Deal. Deal is the professional employer organization, or PEO, that gives you a dedicated HR expert plus Fortune 500 level benefits for your team. The National Association of PEOs says businesses can grow twice as fast if they use one.

16:16So if you're scaling, make it simple with Deal. Go to deel.com slash mos and get up to three months free. Welcome back to Masters of Scale. You can find this conversation and more on the Masters of Scale YouTube channel.

16:39what is the ways you think this will help us also become more creative like what are the kind of creative you know kind of being inspired and inspirational that comes out of the interaction with these agents think about how many random ideas or things that occur to you or questions throughout the whole day if you just really deeply meditate on your subconscious like what of those moments you're like, I wonder, I think, and they're almost like sub-linguistic. They often don't get vocalized because you don't have someone with you all the time to listen to your crazy thoughts other than you. And you certainly don't have the effort to go type something in all the time.

17:17And actually getting your phone out and typing something is just kind of a high bar. Like I would say I search quite a lot, but probably five to eight times a day, it's quite a bit of effort. Yes. So if the barrier to entry to getting those things is now lower, then surely the range of creative thoughts that you can have, which then get manifested in the context of your AI companion has to go up. And then because it remembers, I mean, this is the other big thing that's going to come way before actions is memory. We're going to nail memory. I mean, I'm really confident 2025 memory is done. Permanent memory.

17:53I mean, if you think about it, we already have memory on the Web. We retrieve from the web all the time, quite accurately now. Copilot has really good citations. It's up to date 15 minutes ago, knows what's happened in the news, on the web, and so on. So we're just kind of compressing that to do it for your personal knowledge graph. And then you can sort of add in your own documents and your email and calendar and stuff like that. So memory is going to completely transform these experiences because you will be... It's sort of frustrating to have a meaningful conversation or go on an interesting exploration around some creative idea and then come back three or four or five sessions later and it's like, let's start again.

18:33I've completely forgotten what we talked about. So I think that's going to be a big shift as well because not only does it lower the barrier to entry to you expressing a creative idea, but those things don't get forgotten too. So you can do this ambiguous cross-reference back to something that you wanted. What was that thing I said like three weeks ago? Yeah, and how does this relate to this thing we were talking about? Much more of a conversation. Much more of a conversation. Yeah, exactly. And that is, it's sort of like having a second brain. And it's like an extension of your mind. And that's kind of why the EQ side of it is so important.

19:04A hundred percent. So let's go now down to a little bit more of the tactical with models, because we have a lot of entrepreneurs and kind of thinking about like, okay, here's how to be thinking about how this landscape's evolving in the next couple of years. And, and what are the things to watch for? I mean, the good news is that models are both getting bigger and smaller at the same time. And that's almost certainly going to continue. So there's a sort of new flavor of methods that is starting to come into vogue in the last year, which is known as distillation. You have big, very smart, expensive models that cost a lot for inference, teach small models, and they can do reinforcement learning from AI feedback.

19:41And that supervision, you know, seems to be pretty good. I think there's good evidence of that now. So but scale is definitely going to still be a part of the game. I mean, we've got plenty of room to go. There'll be plenty of data. And so I don't see any sort of slowdown, at least for the next two to three years, in scale models delivering outsized performance. There's also new modalities to put in. So, of course, we're adding sort of video and image and stuff like that. But really what we're sort of, the hard part we're interested in is trajectories of actions across complex sort of digital surfaces.

20:19So jumping from a browser to the desktop and then handing off to your phone and then going from different ecosystems, whether, you know, sort of in your walled garden or in the open web. And you sort of, you know, we're trying to sort of understand these trajectories, collect lots of that data, use supervised fine tuning and so on. And I think that's going to deliver a lot of impressive results. And the other thing, obviously, is there's tons of different angles by which data is talked about a lot. the classic ones, which is like, okay, which data can you run over and what is the quality? I think I'll let that for the, you know, corpus of tons and tons of discussion on the web.

20:54But a little bit of, I think, what people don't spend quite enough time thinking about is where like new data will be. And like, for example, one of the things that I think is interesting about synthetic data is we go, oh, actually, if we had data like this, we could train much better small models, big models. And so how do we get to that data? How do we make sure it's integrated? What's some of the ways that entrepreneurs should be thinking about this? I mean, think about a prompt, not just a question that you ask. I think the language got a bit confused. When you ask a chatbot a question, that's a question.

21:27It's not a prompt. It's a question. When you write a three-page style guide with a set of examples to imitate, that's a prompt. And then you subsequently ask questions of a model that has been prompted. So with that frame in mind, the prompt is kind of your data. It's your high-quality set of instructions which give your pre-trained model direction to behave in a certain way. And it's kind of remarkable that the model can take literally just a few pages of instructions and really behave very differently to a model that has been prompted in a very different way. That in itself is kind of crazy.

22:04But if you just step one step back further, in order for a model to perform with nuance and precision and subtlety and really adhere to the brand values of your business or to the unique product that you're trying to create, you have to show tens of thousands of examples of good behavior. and you have to fine tune those into the model, which is a continuation of the pre-training process with respect to some high quality data that you know to be accurate. Now, the good news is that tens of thousands of examples are very accessible to many niche domains or many specific verticals, right? So that is an edge.

22:45And I think there's plenty of room for startups in really doing high quality fine tuning of a pre-trained model. And then you'll get like much more stable adherence to the behavior policy that you care about. And how should entrepreneurs think about the use and deployment of small models? Because obviously they'll be using Microsoft and OpenAI and Google and other things for the frontier models and scale models to help them with that. Because that's where the multi-billion dollar models will be. But how should entrepreneurs be thinking about like what kinds of opportunities come about with small models?

23:18How can they do something that's interesting and unique with it? Yeah, I mean, I think it's so small is definitely going to be the future. Because if you think about it, the very large model, when you ask a query of a really like, you know, a kind of frontier model, in a way, it's lighting up the neural representations of billions of pathways, which are not relevant to the query at hand, right? And so the crazy thing is it does that incredibly efficiently. I mean, you know, searching or sort of referencing like hundreds of millions of nodes, if you like, at each token that is generated. You know, that's kind of crazy.

24:00But it doesn't need to do that. If you have a tight use case, then what I think is going to happen is we're going to sort of compress knowledge into smaller, cheaper models which can live on a fridge magnet, right? And, you know. I haven't heard you use that metaphor before. A fridge magnet? Yes. I don't know what kind of the smallest digital thing I can think of. I'm not even sure it's digital. Oh, right. Yeah, that's true. Yes. Yeah. Like it welcomes you in the morning and it'll talk to you. Or like it's going to be on an earbud or a wearable or my earring or, you know, in a plant pot with a little sensor or, you know.

24:40So those things, like the ambient sensing revolution is going to come alive, I think. It's been sort of long promised. but that's the kind of compression trajectory it'll go to the extreme where you can have really quite functional you know obviously the fridge magnet is not going to know a lot about quantum computing but it's going to know what it needs to know in order to welcome you in the morning give you the weather talk about what may or may not be in the fridge yes remind you of your your calendar that i think is going to be maybe a few tens of millions of parameters yes and we We haven't, you know, people aren't really pushing that yet.

25:14And yet it's totally feasible for any, you know, two-person team to go and explore that. Yeah, exactly. And that's part of the sense that this is also an entrepreneurship event is a key thing. So I'm going to move to kind of a slightly longer version of our last question, which is what's the question that people should be thinking about for the next two days? And to give you a bit of, you know, a few seconds to think about it, having sprung this upon you. I'll start, right? And, you know, for me, that would be, and I'll generalize off what I was just saying, but it's like the, what are the things that we need to bring in as technologists to be thinking about how do we design a more human future?