The Changing Physics of Cybersecurity and Drug Development with OpenAI President Greg Brockman

Technology has always been the fundamental driver of improvement for human quality of life. And with the recent rollout of advanced frontier models, like OpenAI’s GPT-6, humanity can now massively accelerate scientific discovery for its benefit. Sarah Guo and Elad Gil are joined by OpenAI President and co-founder Greg Brockman to discuss the company’s rapid progress across frontier models, cybersecurity, and scientific discovery. Greg shares how nearly 400 century-old math problems have been solved using its AI agents, and how the technology has the potential to transform other fields like biology in the realm of drug discovery. He also talks about how proactive, always-on AI agents like Dots impact modern productivity, why he believes the best UI is no UI at all, how cybersecurity is in a critical window of time to use peak intelligence to find and fix vulnerabilities in systems, and why OpenAI remains committed to keeping humans in the driver’s seat even as models surpass us in capability. Sign up for new podcasts every week. Email feedback to [email protected] Follow us on Twitter: @NoPriorsPod | @Saranormous | @EladGil | @gdb | @OpenAI Chapters: 00:42 – Trailer 00:59 – Greg Brockman Introduction 01:11 – AI and Advancements in Mathematics 05:55 – AI Communication and Oversight 07:45 – AI for Overcoming Biology Data Bottlenecks 11:28 – Revolutionizing Drug Discovery 14:18 – Changing Regulations to Accelerate Progress 16:51 – Dots the Proactive Agent 19:19 – Unifying Work and Consumer Agents 22:02 – How UI Will Evolve 26:08 – Cybersecurity in the AI Age 32:55 – Shift in Security Investment 35:49 – OpenAI’s Human-Centric Ethos 38:28 – Conclusion

• 0:00Speaker 00

  To ensure AGI benefits all of humanity. We think that humans should be firmly in the driver's seat, should have control over the future, and this technology should amplify human agency. When I look at the world we're going to be in, where software, in some ways, is becoming like the barrier to entry, it's becoming so cheap. You don't just want software, right? You want software that is reliable, that is secure, that you can believe in. And so cyber spend, security spend, should actually vastly outstrip software spend. I think that people have not yet really internalized how much the physics around cybersecurity are going to change. I think there's an opportunity to end up with a much more secure world.

• 0:39Speaker 01

  Hi, listeners. Welcome back to No Priors. Today, Elad and I talk with Greg Brockman, the co-founder and president of OpenAI. We discuss OpenAI's latest math results, Astra, and why Greg believes AI can now generate genuinely new knowledge. We get into drug discovery, DOTS, and his case for why security spending should actually vastly outstrip software spending, which is quite a claim. Here's our conversation with Greg Brockman. Greg, thanks so much for doing this.

• 1:06Speaker 02

  Thanks for joining us.

• 1:07Speaker 00

  Thank you for having me.

• 1:08Speaker 02

  So every week in AI feels like a year, and a lot's happened this week for OpenAI. You rolled out GPT-6 to over a billion people, which is pretty amazing. There's some really exciting news that you have on the math side. There's all sorts of things undoubtedly that are happening internally. What are you most excited about that's happening right now, both at OpenAI and in the broader AI world?

• 1:29Speaker 00

  Look, I think that in AI, like I have this belief about how OpenAI operates, that there's this cyclone quote of it never gets easier, you just go faster. And I think that's very much what happens, is that there's just so much activity and that every time you think you're at kind of peak, sort of, you know, impact and how things are happening, you realize that there's actually another level that's possible. And it's not just about any one company. It's not just about any one technology. It's really about we, as a world and society, are all coming together to adopt this technology, get the most out of it for everyone's lives. And so I think, to me, the math results this week represent a real milestone. Right? It's a real moment where the, if you think about like the fundamental driver

• 2:12Speaker 00

  of human quality of life improving over the past thousand years is technology. Right? It's scientific advancement. And I think that we now have the ability to massively accelerate scientific discovery. And there's always been this question of, is AI actually able to come with new knowledge? And at this point, it's just impossible to deny. Right? It is definitively yes, possible. Now, there's still more to go. I think we're not yet coming up with whole, grand new theories, but we are able to resolve these hundred-year-old math problems. And to me, the thing that's very exciting is it's not just about the sort of, you know, sort of the just, the machine just sort of gives you an answer, right? It's really what can a human equipped with these tools achieve that was unthinkable before?

• 2:58Speaker 00

  And I think it's almost like that the leverage of humans to do whatever you want to do, it's going way, way up. Right? It's now that you can, if you have a math problem you want to go solve, you can solve it. But what about taking that same capability, applying it to biology? What about if you're a small business owner and you want to be able to operate your business well? Like that level of intelligence that's tuned for that domain, it's just going to mean that almost this sort of unthinkable results that are aligned with your goals, with your desires, is now going to be within the breach of everyone. And that, to me, is so exciting.

• 3:32Speaker 02

  The rate of progress has been pretty amazing, too, because if I look at it, you know, you used to have problems with arithmetic, right? In early models, I remember people complain, oh, why can't it do basic math? And now you're solving these advanced theorems. As these things continue to advance and this starts to really impact areas of math, physics, materials, et cetera, and various applications, is there some need, if these models get so smart, to focus on how to communicate that effectively to people? Or how do you think about the actual communication side of this as you do a very advanced theorem? Because if you look at any field of math, only a very small number of people can ever really adjudicate a theorem that's produced by a person.

• 4:07Speaker 00

  Yes.

• 4:08Speaker 02

  Right? And so you kind of have this specialization that happens.

• 4:10Speaker 00

  Well, and this, by the way, is something that has always been, like I came from a math background. Like I thought I was going to be a mathematician. And I was very inspired by mathematicians like Galois or Gauss, right? These people who are working on like a hundred, four-hundred-year time horizon. Right? And I think there's something deeply beautiful about mathematics being about sort of understanding the structure of the universe. And so it's not just about, and there's always these questions about is math invented or discovered? And that, in my mind, it's something that the beauty and the meaning of mathematics, it's not something that is objective in some sense. It's in the eye of the beholder. It's something that appeals to us. And there is some amount that's like kind of physical laws

• 4:54Speaker 00

  and it's about simplicity and, you know, compression theory. You know, it's just complexity theory and all these things like that are maybe are more physics, but there's something that is also importantly human about it. And so to me, it's not just about, and I think a lot of the conversation about mathematics, it's not just about here's a proof, no one understands it. Great, we resolved the problem. It's really about what are the insights that are derived? And the thing that I think we will see, but I think that there's promising signs of life on, is that there's beautiful insights within the work that this, that these AIs have done. And I think what we're seeing right now is that,

• 5:33Speaker 00

  you know, we published about almost 400 very important open problems that are now resolved. And the mathematical community is going through all of that to look for insights. And if you read some of the things that people have posted, you know, domain experts in these particular problems, and they say that it's like that there's these almost like sort of very surprising ways of applying ideas or incredibly like, wait, why did they use that result over here? And that what's coming out is not these incomprehensible theorem or proofs, which, by the way, there is a history of incomprehensible proofs in mathematics. Like four-color theorem is a perfect example of this that is, it's like, okay, like we just had the machine check a bunch of cases. Like there's no beauty in that.

• 6:18Speaker 00

  And so I think we're into a different phase of machine discovery.

• 6:21Speaker 01

  I have some optimism that like as models also get to better theory of mind, like the ability to figure out like what we as humans know and how to get us to follow them on the path to new understanding will also be paved, right? And so like, for example, like I'm a math person, but at some point in school, I stopped learning math. And then I did Math Academy. And then I did like ChatGPT to help me learn new math. And so I just feel like the expectation for, you know, the rising tide of what knowledge, like you can expect people to understand will also be there.

• 6:52Speaker 00

  I think that this is a really important point and one of maybe the underappreciated benefits that I expect from and hope for and are starting to see from this technology is helping people know more, understand more, understand each other better, understand the world better. And I think that there are technical mechanisms that can underlie the, you have an AI that's able to go and solve these problems, see all these connections between domains. But how does it actually communicate it to you? How do you understand it? And actually, there's even connections to AI safety and alignment on having AI that we're able to understand what they're doing. And so there are techniques that we've been talking about since 2017, such as debate and iterative amplification, which are kind of two sides of a similar coin,

• 7:35Speaker 00

  where you really think about if you have AIs that are solving problems that are just too hard for humans to check. And even, for example, summarizing a book is an example of a task that's really hard to check. Right? It's like, it's really hard to know is the summary any good unless you've read the whole book? And so if an AI just gives you a summary of a book and says, this is the summary. Is it a good summary or not a good summary? It's pretty hard unless you have another AI who is trying to prove that the summary is a bad summary. And so then if that AI comes back and says, hey, like, here's this like paragraph in the book that I found. It has, it's got a clearly a very important plot point, and that is not representing the summary. You don't have to read the whole book to be like, oh, yeah, that's actually a very good argument. And so you realize that there are ways to train the AI

• 8:17Speaker 00

  that are actually very aligned with having humans be able to have oversight, to be able to be in the driver's seat, which is fundamental value, a fundamental thing that is important for the future from OpenAI's perspective. But also are very much aligned with what you're saying of, can you make the results of an AI be something that means you learn something that is really tuned for the human, rather than just sort of knowledge for, you know, some abstract knowledge sake?

• 8:42Speaker 01

  For most people, advancements in math theory are still somewhat abstract. Right? So that there's like, wow, this is absolutely happening today. And then also, you know, AI for science, like there's a claim or a belief that it could solve all disease in some number of future years. Right? How do you look forward like one step in terms of what we might see in advancements in between?

• 9:05Speaker 00

  Well, I think that we've had a number of years in iteration of this technology to see the shape of what to expect. And I think that it's kind of like when you have an AI that is capable in a certain dimension, and if you understand the like, why it got capable in that dimension, then all other fields of human endeavor that have similar properties, you should also expect similar results. And so in my mind, like, why is mathematics something that we've managed to really transform in the way that we have had such incredible results? And that part of it is that our AI is now at a certain level of capability and raw intelligence. Part of it is that You could think of it as that, well, mathematics is checkable, but it's not quite true anymore, right?

• 9:49Speaker 00

  Like, we really started with very checkable problems of, here's an arithmetic problem, and there's like some numeric answer, and so you have an AI that you can train to just like, check if it got the answer right, because we can very much deceive is the answer that the AI is proposing the same as the actual answer. But for a proof, very different. Proofs are very hard to check. Is this actually accurate or not? Formalized proof, okay, yes, that is checkable, but, but natural language proof, which is what we're really producing here, much, much more difficult. But the thing that actually I think is the core insight is really about data, right? It's that you can get lots of data in mathematics, you get lots of data in theoretical physics. Can you get lots of data in biology? That's obviously something very data bottleneck, very difficult.

• 10:31Speaker 00

  Now, fortunately, we also have been working on lots of techniques to be able to work with less data. And if you think about what that means, it's like, okay, we now have smart models that if there's something where there's a lot of data, and this is the fundamental physics of deep learning, right? Compute data algorithms equals capability equals results. We're able to do it in these hard domains where people have spent hundreds of years of our smartest people all trying to solve the problem, and now the AI can solve it. Amazing. When you transfer that over to biology, and clearly what we want is drug discovery, all these things that are like fundamental changes in biology and our understanding of it to improve quality of life, that's what we want to apply. The question will be, and is not yet resolved, do we have the data?

• 11:16Speaker 00

  Do we have the ability to get the data? And so I think that what will happen is that we will take these, and we're already doing this, where we take these, these models, these same techniques, and apply them in the biological domain, and there are some areas that will fall quickly, right? That we'll just see the massive acceleration of things that used to be incredibly painful, now incredibly easy, but there will be other bottlenecks. And there'll be a question of, well, how do we alleviate those bottlenecks? And some of these, by the way, are not even really about deep biological theory. For example, regulatory filings, huge bottleneck, right? Takes like a year to write up your whole sort of case for clinical trial and all these things, and it's just like, you could write those things with an AI much more easily. And so I think that we're going to have to look at every step of the drug discovery

• 11:58Speaker 00

  process, and there are going to be places where the fundamental data just does not exist. And so then innovations on wet labs, innovation on how you actually do simulations, all of this will become incredibly important. But, so I think that we will see these breakthroughs coming, but it's important to know there are things that are shared with mathematics, that is about scientific discovery, and things that are very different in terms of where the data comes from.

• 12:23Speaker 02

  There are specific, you mentioned biology, and I think that's a great example, where a lot of the early research, alpha fold and things like that, were really focused on designing a molecule or understanding how something binds to each other. And to your point, it takes 15 years to develop a drug, it's $1.5 billion. 15 million or so tends to be preclinical, mouse studies, the molecule, and then 1.4 billion plus tends to be the clinical trial. And so your point is, and it's also, you know, two, three years versus another 12 years. And so your point is really well taken in terms of a lot of the early emphasis of these models has been, can the thing bind versus what are all the other things you have to worry about downstream to actually turn it into a productive thing for human endeavors? How do you at OpenAI currently stack rank the different areas of scientific endeavor

• 13:05Speaker 02

  in terms of where you're focused, or where you wanna generate more data, or more results, or more capabilities that others can benefit from?

• 13:11Speaker 00

  Look, I think that drug discovery is the most clear and obvious area, if you get to pick one, that you want to fully revolutionize. And just in places where you're going to really uplift everyone's quality of life, that is just like so purely important, and maybe the top one application of AI that everyone's been kind of thinking about for a long time, we clearly got to do it. And to your point, like, there's this whole pipeline, it's just so long. And I think we're going to find that humans are incredibly bad at biology, incredibly bad at drug discovery. And if you think about it, there's this like property where for humans, natural language, we're very well suited to it. Like, it's something that, you know, very clearly, it's very human, it's like very

• 13:56Speaker 00

  well tuned to, to how we have interacted with the world, whereas biology, it's totally foreign. But for a neural net, I don't really think there's any reason to believe it should be better at natural language than it should be better, you know, compared to biology. And so the fact we're seeing such incredible results with neural nets and natural language, you're just like, from first principles, you should be able to really be able to crack the language of biology. And so to me, that is number one. It's like first principles, it should work in terms of what we're seeing, we're so close on so many acts.

• 14:25Speaker 02

  Yeah, it's very exciting.

• 14:26Speaker 00

  Yeah, and I think you can massively increase the parallelism. I think you can increase, I think you can decrease so many of these steps in terms of, of their duration. And I think, by the way, that there are, there are many dimensions to drug discovery, for example, N of one diseases, there's even like lots of people who the drug to cure their particular condition exists. It's just that, it's just out there somewhere in the literature that there's a paper, and if you can find that one paper that is relevant to your conditions, then you'll be able to cure the person. And this, there's plenty of precedent for it, and it can massively scale up those kinds of operations. So I think that impact in the short run is actually very possible. But material science, another example, and I think there's a long tail of, of other domains where we should just see real changes in, in kind of what is possible for humanity

• 15:11Speaker 00

  to accomplish.

• 15:12Speaker 01

  It seems, just, we think about why a drug might take 15 years to go to market. Part of it is you're doing like successively more expensive trials that have a, you know, different set of safety considerations resulting in large scale human trials. And, and then, these trials today are, they're ordained by regulation, right? For a good reason. It feels like one of the limiting factors, if you believe the technology will work and parts of the data can be acquired or generated, is going to be convincing regulators that we should change the expectations around, around what drugs are accepted and what the steps are.

• 15:51Speaker 00

  I think there is a moment here where, and there's more work to be done to reach this point, but if you can really prove that the physics of drug discovery have changed, then of course it should trickle through to every part of the system. And if you think about, I mean, even just take a very different example, think about self-driving cars, right? We have so much of our, how roads operate and safety and all of these things that are really tuned for human drivers. And if you have, if you reach a point where self-driving cars are 10 times or 100 times safer than, than human ones, right? That you've reached that point where AI is actually able to deliver the results in a domain that just are so unambiguously the thing that you want, of course you should

• 16:33Speaker 00

  then say, like, okay, how do we rethink the whole transportation system? Maybe you should have regulation that says that you only are allowed to have self-driving cars in certain areas, like, that there are a lot of, of potential, sort of, changes to the, to the, the way that you think about what is even trying to come out of a system that needs to trickle back into regulation as well. And of course, regulation is not uniform across the world, right? That there's very different regulations in Australia versus the US when it comes to clinical trials, and that you're already seeing a lot of, of, of people moving their, their trials to places where the regulatory environment makes more sense for how, how things are, how they want to pursue, and, and, you know, what people think is, is efficacious for, for them.

• 17:13Speaker 00

  And so I think that the, exactly how things should change, I think will depend a lot on what we actually can prove, right? Is it the case you actually reached the point that you can get better results from a neural net than from actually running a clinical trial? Like, we're not there.

• 17:28Speaker 01

  Yeah.

• 17:29Speaker 00

  But you could imagine we will get there.

• 17:30Speaker 01

  But if they were more predictive on efficacy or tox or something, then obviously you should change it.

• 17:35Speaker 00

  That, of course. And I think that of areas, of areas where I think society should, and I hope will, be very sort of leaned into the, what is possible has fundamentally changed, and let's very quickly then reorient around it. And again, you have to do it, you have to really think through all the, all the consequences, right? And all of the implications, and that, one of the, the, you know, if you're thinking about long-term studies, people and like question of like, you know, is this neural net actually, in terms of what's predicting the simulation, actually accurate for long-term outcomes? It's pretty hard to not have at least had, you know, one long time tick to, to correspond to that.

• 18:15Speaker 00

  But I think that, that, I think we will see like 80% of what is possible come to fruition very quickly.

• 18:24Speaker 01

  You're a little ways out from the Astro release and Dev Day now. Is there anything in terms of how people internally or externally are using some of the new products or capabilities that surprises you?

• 18:38Speaker 00

  Well, I think Dots has really changed how many of us work. And the thing that Dots represents is-

• 18:44Speaker 01

  Can you describe Dots briefly?

• 18:45Speaker 00

  Yes. So Dots is a proactive agent, it's always on, always working. You stop working, it keeps working. And the kind of thing that my Dots does, right? So I have, I have a dot that's looking out for work that I was supposed to do, like messages I was supposed to send, things I said I would do that I haven't gotten to yet. And I always wake up in the morning, there's like a list of things that I was like, oh yeah, this is my to-do list, and that I, it often notices things that I never even asked it to notice. If there's like a conflict on my calendar in terms of scheduling, it'll point that out. If it's like, hey, you actually need to leave this meeting earlier in order to get to the other thing. And so it's like so contextually aware. It knows what I'm doing with my day, it understands the projects we're working on,

• 19:27Speaker 00

  it understands, I sort of, you know, it knows what's in my inbox, knows what's in my Slack, and is able to integrate across these. And it knows the questions I've been asking about, so it knows what areas are most of top of mind for me. And I think that having, again, some ways, what Dots is, is to change to the form factor, not the capability, right? It's like, it's an Astro level model, it's got its own cloud computer. But that the change in that form factor in a workplace setting with this level of intelligence, it's a game changer.

• 19:56Speaker 02

  And do you have multiple dots running or you just have one that's sort of covering all your background tasks?

• 20:00Speaker 00

  I've got one, I've got one. It's a little owl is the icon. You can ask it to change the icon. And it's just sort of doing everything across my work life.

• 20:13Speaker 01

  OpenAI obviously has, you know, research efforts and then a consumer and an enterprise and developer business, platform business today. Do you think that there's going to be a separation between, let's say, assistants that help you at work, like dots, which is how I primarily think of it, and then like consumer assistants like the Instincts Muses of the world?

• 20:32Speaker 00

  There is some, I think these things are going to converge, no question. And I think that there are some pressures that are very different in different environments. So for a mass market consumer product, really figuring out how to scale and to provide things for free or at very low cost is very important. Whereas in the workplace and if you're productivity, if you're the executive, individual contributor, whatever your role is, then it is much easier to say peak intelligence and really, as long as you're getting returns to paying more, it's worthwhile to throw more compute at it. And so I think that that is the main distinction is really around cost. And we've seen this in ChatGPT, right? That ChatGPT, we put in a lot of effort to have the largest scale free tier

• 21:17Speaker 00

  and bringing things like memory to all of our free users was an incredible endeavor, like a real huge engineering effort. And it's just very easy to implement the same functionality at much higher cost for a very small number of people. And so I think that what we're seeing in the personal assistant workspace is that there is, there are going to be these different market segments where different amounts of compute are going to be expended fundamentally. And so sometimes when there's different levels of compute, it then means the product can look a little different, right? It can be that there's just more functionality that's embedded in one versus the other. And there's of course, different use cases when you think about things like, you know, payments and travel and things like that are much more for the consumer. But I don't think there's any fundamental difference, right?

• 22:01Speaker 00

  It's like really just like an AI that's doing things and it's just like, happens to be like slight sort of pressure in one way or another. And so one of the strengths of OpenAI is the fact that we have 1.2 billion consumers on ChatGPT. And we have also a leading coding and enterprise business and no one else has both in-house. And so that a lot of our strategies is to say, let's unify these. So we're not sort of, you know, sort of competing on two fronts. It's really competing on one front. And I think that this is also the general direction of travel of AI is to blur the distinction between what is consumer? What is enterprise? We're going to have this renaissance of entrepreneurs, like we're already seeing it. I mean, imagine you're seeing this very concretely. I know people who are quitting their job to go start companies who say, I never would have done it

• 22:45Speaker 00

  if I didn't have an AI that I knew I could rely on, it could take care of so much of the normal barriers to entry. And so I think that we're going to see, just like the distinctions have held for so long, I think they're going to change.

• 22:56Speaker 02

  So one of the things that traditionally has separated the enterprise and consumer world is how you think about UI. Obviously, there's your point on cost and, you know, underlying ability to deliver something for somebody at a certain level of fidelity. But the other part of it is just the user interface. And often the user interface in an enterprise setting or how you access it is very different from the consumer setting. To your point, you're looking towards unifying these layers. You've also been thinking quite a bit about like dynamic UIs. And that's always been kind of, one of the things people have talked about quite early is how do you just dynamically generate a UI that's useful for a user? Or in some cases, these things kind of move into the background as just part of your chat. How do you think UI will evolve for these different use cases? Or how do you think about that as a problem as you sort of address both of these segments?

• 23:37Speaker 00

  So first of all, I think that the best UI is no UI. And I think in AI, that is where we're going. Just one thing you can talk to and you can speak to it verbally, you can talk to it over Slack, iMessage, any of the surface you can call it. And that's literally what you can do with dots, like literally today.

• 23:57Speaker 00

  But there are some fundamental differences between personal setting and a workplace setting. And workplace, you have an admin, right? You have kind of your company security team. You have just more stakeholders and it's not just sort of a AI that's for you, that's representing you, that you own and you own its outcomes. And so I think that that is the main distinction I see between the enterprise and the home use case. And of course, this leads to product differences as well. And sometimes there are very small things, like for example, when we release a new feature, that enterprises want control over when it's introduced. And so we have admin controls and toggles and so you can have an experience

• 24:42Speaker 00

  where there's just less functionality available to you as a result of the fact that it is in the workplace. And that does then change the, well, we need somebody to surface that to you. And this comes back to, well, the whole point was that there was no UI, right? So it's like, what does it mean for there to be functionality that's missing? How do you even surface it and let people know that that's the case? So that's an ongoing challenge and I think that that will not magically go away. I think that you need to lean into that complexity. But yeah, to me fundamentally, it's kind of crazy and beautiful if you think about it. The ChatGPT, the text box, right? That's where it started, it was a text box. And what we're heading to, there isn't even a text box, right? It's just like, that is the AI we were promised.

• 25:20Speaker 01

  When I think about what might make a product that doesn't have a UI better, there's like, you tell me, but I think there's like, how much compute can I spend on it? How good is the model itself? There's a memory system, there's like, I don't know, decisions about behavior. How good are models that being calibrated on, you know, how hard they should think about something?

• 25:44Speaker 00

  Is that something we're getting very good at? And with the release of GPT-6 to everyone, that we have the ability now for the AI, if it's going to go think hard about a problem, it actually can give you a quick answer and then keep thinking and then give you the more fulsome answer. And also to the point of no UI, and to the thing you were saying earlier too, is that we introduced intelligent UI. So now the AI can do generative UI on the fly. And so I think that even though there's no fixed UI, there's no buttons, there's no menus to learn, having an AI that can actually craft the result to be visually exactly what you want. Like you think about even something like flight booking, you want kind of a complex UI there, right? So you can see all the different options. And that for any question you might ask, like how do I assemble a bicycle or something?

• 26:28Speaker 00

  You might want some very sort of bespoke app that normally a person would have gone off and written for a month. And of course, you're never going to hire someone to write a one-off app for you to help assemble a bike. But totally fine for an AI to go do that for you in real time, in the moment you need it. And so I think that there is something about these, like we're now in a world where the AI is molding itself much more to you, right? That the user interface is becoming exactly the thing you need at the moment you need it. And that is a lot of our focus and we're doing that across every aspect of how we think about the product.

• 27:02Speaker 02

  One of the biggest shifts in the discourse around AI recently has been around cyber. And how do you build secure systems that are resistant to AI-driven cyber approaches? And so I'm just a little bit curious how you're thinking about that? How are you building guardrails or how you're approaching the different aspects of cybersecurity?

• 27:18Speaker 00

  Well, first of all, I do think the physics of cybersecurity are changing. And we're in what we call the defender's window. And the way that I think about this window is that we have peak frontier AI capabilities that can help defenders secure themselves. They're not broadly diffused yet, but that the broadly diffused capabilities are ratcheting. And you can see this from each model release to model release from providers across the world.

• 27:43Speaker 00

  The capabilities on cyber that are available to everyone are going up. And you're starting to see the leading edges of some of the impacts, right? You're starting to see, for example, in South Korea, that there were some high-profile hacks of banks that were driven by different models. And I think that the picture that is so critical is that defenders need to use this time. And so we have been working very hard to be able to responsibly give access through our Daybreak program to defenders to be able to use our peak intelligence or to use very capable models to find and fix vulnerabilities in their systems. And one thing that is very important to realize is these flaws have existed, right?

• 28:25Speaker 00

  It's every line of human-written code in some ways that you have to ask, is this secure? Was the developer trained in the right way? And you just start to realize that the way in which we built systems were kind of fundamentally insecure. And we always knew it, right? There's never been a CISO who's been over-resourced.

• 28:44Speaker 01

  Yeah. I mean, you can look at X KCD and the comic of like all the world's infrastructure reliant on like one open source project written by one dude in his basement, right? Or Heartbleed or anything that predated like all these AI issues.

• 28:56Speaker 00

  Exactly. And so I think that to me, and I think very importantly, it's very important to realize that these are really human issues, right? That these are, the way that we chose to build the internet, it was, security got kind of bolted on, right? And we had high-profile incidents, right? You think about early worms and viruses. And then people started to realize, well, we need virus scanning software, right? And we need to have encryption for requests, you know, over HTTP should be over HTTPS. All of these mechanisms were ones where I think that we kind of took the slower route in order to add security on afterwards. Like the internet's really been built for resilience, not necessarily for security.

• 29:37Speaker 00

  And I think that what we're going to see now is a moment where a lot of that debt, a lot of that tech debt is going to come to the surface. It's time for us to pay it down. And that as a company, we are doing, we are really leaning into trying to help with this moment. So for example, we have a billion-dollar commitment that we've made to help frontline defenders get access to our models and utilize them for defense, that we're studying and trying to find different vulnerabilities out there. And then thinking about how do we help the world respond? How do we coordinate incident response? How do we work with government? How do we work with the community in order to land these as well as possible and to help people navigate through?

• 30:17Speaker 00

  But I think that people have not yet really internalized how much the physics around cybersecurity are going to change. I think we will end up, I think there's an opportunity to end up with a much more secure world, right? You think about the fact that AI lets you pay down your tech debt in a way that was just previously unthinkable. The fact that you could do things like formal verification of software. Like these models are obviously great at proving things. Why not prove security properties of software? That's something that the technology has kind of been there for people, but it's just so much work.

• 30:46Speaker 01

  Painful, niche, expensive.

• 30:47Speaker 00

  Exactly. No one does it for anything significant. And so I think all of that can change. And I think the world we're moving to right now, and we've done this internally within OpenAI, is every time a new model drops, you need to apply it to secure yourself, find your vulnerabilities, and then saturate. And so, for example, we took about 25% of our production engineers and said, "All of your projects are in hold. You are just securing now." And took our peak intelligence model at the time, had it poke at all of our infrastructure, rating a bunch of things, built some tools so that it's easy to do it again over time. And we found a number of issues, fixed it. And at some point, we saturated the issues that this model was capable of finding. And that's a very good story, right? It's like, I think that the thing to realize is that we're going to have a next model.

• 31:32Speaker 00

  We're going to have to repeat that process and that that will continue. But at some point, again, I think if you think about, well, the code that's being written, we're actually, so much of it is now AI assisted. We can actually train these models, and we are training these models to write secure code from the beginning. And so I think we can have sort of fundamentally more secure systems through the use and application of these AI models in the right way.

• 31:54Speaker 02

  So it sounds like basically there's going to be a period of time where maybe we're going to see more of these cyber attacks, more incidents, et cetera, as humanity or as people sort of catch up in terms of their core bases getting more and more secure, using more AI-generated code that's sort of preset to be more secure. And so we'll probably end up with some transitionary period over time where, you know, we'll have more offense and defense, and then defense will catch up. I mean, that's kind of the consensus I've been hearing in the industry in general. And it seems like the most likely forward case, at least.

• 32:22Speaker 00

  Yeah, look, I have a lot of optimism about how we can navigate this moment. I think it does require a coordinated response, right? It really requires defenders to internalize how the physics of cybersecurity are changing, that there is this window where that window is not going to last forever, right? And that you should think about, we're shifting to this world where the security of our organization is related to both the fundamentals that they've gotten right, which is kind of a one-time investment, but also the level of cyber capability they've used to defend themselves. And as we shift to that world, I think that we can be in a world where defenders are able to outstrip attackers.

• 33:04Speaker 00

  But it is going to take a lot of change, right? It's going to take a sort of change in terms of how we think about every organization operating and how we collectively think about critical infrastructure and resourcing those that have sort of not had those resources to date.

• 33:21Speaker 01

  I think your example, I mean, your description of this is like, let's say, like, internet-wide tech debt, and then your response to it internally is a really good one. Because I'm like, okay, I've been part of companies where, you know, tech debt is very painful to address and to say, "We're going to take 25% of the engineers," or, God forbid, you know, "We're going to do a feature freeze for a quarter until we fix some stuff that really needs to be fixed." That is, that doesn't feel like the state that the world has decided on quite yet. When you think about where the entire world is from a security investment perspective, traditionally it's sort of this like, oh, maybe it's like low, single digits percent of the IT budget. Like, what do you think it should be at this moment in time?

• 34:04Speaker 01

  And like, is it all of this other stuff that you were investing in from a cyber tooling perspective, network endpoint, traditional corporate security is less relevant? Or we just need to go spend twice as much?

• 34:16Speaker 00

  I have a feeling that cyber spend, security spend, should actually vastly outstrip software spend. Like, I really think that especially--

• 34:25Speaker 01

  Well, that's a statement.

• 34:26Speaker 00

  Yeah, I really believe it. And I may not be right, but when I look at the world we're going to be in, where software in some ways is becoming like the barrier to entry-- Exactly.

• 34:37Speaker 01

  Free-ish.

• 34:38Speaker 00

  Exactly. It's becoming so cheap.

• 34:40Speaker 01

  Yeah.

• 34:40Speaker 00

  And so, of course, making sure, you don't just want software, right? You want software that is reliable, that is secure, that you can believe in. And so, to me, the resilience of that software is the sort of--

• 34:53Speaker 01

  That's the value.

• 34:54Speaker 00

  That's exactly, that's going to become even more important. So I think that we have this moment, we have this opportunity to, you know, to your point of the XKCD with the little block at the bottom of some open source maintainer--

• 35:03Speaker 01

  Yeah.

• 35:03Speaker 00

  Like, that never made any sense. And the reason we had that world was because of some of this physics, right? Of kind of how hard it is to write software and to write reliable software and that everyone has the supply chain. And so few companies audit their supply chain, right? If there's, we've seen this with supply chain vulnerabilities where someone uploads a malicious package and a number of companies get hit. All of that needs to change, and it was totally impossible to change with human software developers because there's so few, there's so few people who have this expertise in this very obscure thing of like, you know, sort of knowing different packages and like different security techniques.

• 35:38Speaker 02

  Third-party software now, like Socket and others, who'll do this as an example, where they'll scan through your open source supply chain and try and find vulnerabilities and identify them for you. Or, you know, there's the human element of it where maintainers would suddenly take over a project and then install something that would be malicious relative to people than using that packet. So it's really interesting to see that shift in terms of the third-party ecosystem as well, developing to support the efforts that OpenAI and others are doing in terms of this market.

• 36:06Speaker 00

  Exactly. And I think it's so important. So we're doing this with the community. We're partnering with companies like CrowdStrike. And I think that it is, again, a coordinated response, but we really view it as a core priority for OpenAI to lead, right? For us to really, because we have the frontier model, we have the peak intelligence model, certainly in cybersecurity, we think that there is transformative potential for the good possible. But that means that we, as a company, and we've been going through a security transition, right? That over the past couple of months, we've talked about safety, security alignment of our development process, but we're also thinking about that for the world of how to really help the world become more fundamentally secure and help people through this whole period.

• 36:44Speaker 02

  One of the things about OpenAI is that it's always been very pro-human and very pro-human agency, human progress, let's use AI to better everyone. I think the original premise of OpenAI was, let's have something that can counterbalance other companies in terms of creating AI that can be used very broadly and universally in a positive way. Has that ethos continued at OpenAI? Or how do you think about that in the context of the AI community?

• 37:05Speaker 00

  I think that ethos is very core to our mission, right? To ensure AGI benefits all of humanity. And the humanity there is very key. We think that humans should be firmly in the driver's seat, should have control over the future, and that this technology should amplify human agency.

• 37:19Speaker 01

  Even as models surpass humans in capability.

• 37:22Speaker 00

  And I think this is absolutely critical, right? That I think that the picture of, and keep in mind, these models are very jagged, right? That there are places where obviously these AIs are more capable than anyone today, right? Even GPT-4 knows far more than any person. But the thing that is so important, and I'm reminded, for example, of the, there's the story of why Steve Jobs put the handle on the iMac, right? So that that way, that people could feel that they could control the machine, right? That you can always pick it up and throw it out the window. And I think that the same spirit needs to apply to AI. The fact of the spirit of humans are in the driver's seat, that these, like, human judgment, human values.

• 38:05Speaker 00

  And maybe you should think of how humans will operate in certain fields, in certain dimensions, in the future, whatever it is, is we become more like CEOs or like board members. But it's important that we have that level of agency and that I think that we're going to build machines, just like, you know, if you're a CEO, you're clearly going to have domain experts, you're clearly going to have teams that are going to have knowledge across so many things and maybe even better ideas than you on many different access. But fundamentally, the question is, are you happy? Is this what you wanted? Is this what makes your life better? And that, to me, is something that we should all collectively-- it's not just about AI, it's about technology, right? Technology is to lift up human civilization.

• 38:50Speaker 00

  It's the single most fundamental driver of improvement in quality of life. And that is what we should be seeking from AI. That is the belief of OpenAI.

• 38:58Speaker 01

  Thank you so much for doing this, Greg.

• 38:59Speaker 02

  Yeah, thanks for your time.

• 39:01Speaker 00

  Thank you both. Appreciate it.

• 39:04Speaker 01

  Find us on Twitter at NoPriorsPod. Subscribe to our YouTube channel if you want to see our faces. Follow the show on Apple Podcasts, Spotify, or wherever you listen. That way you get a new episode every week. And sign up for emails or find transcripts for every episode at no-priors.com.