No Priors — Elad Gil and co-host with Greg Brockman (OpenAI): SI for math, biology, agents, UI and cyber — denser notes · Don't Panic Digest More
What this is: Nearer-verbatim notes from a public X interview video. Not a court transcript. Gaps labeled. Disfluencies lightly cleaned. Hosting ≠ endorsement.
Not Code Solutions LLC viewpoints or advice. Speakers' claims stay speakers' claims ([unverified] / host-said).
Hedge legend: [unclear] · [seam … filled from audio check] · [unverified] for claims and figures · host-said for host framing · "Host 2" for the co-host, whose name is not spoken on the audio.
Speaker claims: Claims are the speakers' own [unverified] — not checked, not restated as fact. Greg Brockman's views are his attributed opinions. Examples: that SI "can now generate genuinely new knowledge"; the published math results and the count of resolved problems; product rollout reach; the drug-development cost and time figures Elad Gil cites ($1.5 billion, $15 million preclinical, $1.4 billion plus clinical); "1.2 billion consumers"; South Korea bank hacks "driven by different models"; the Daybreak program and a commitment for defenders (amount left out; audio passes disagreed); "about 25% of our production engineers" moved to security; and that security spend "should actually vastly outstrip software spend." Timestamps are approximate (player time); within a turn they mark the nearest preceding cue.
Source URL: https://x.com/eladgil/status/2110812265238991097
Full video duration: 39:22 (native X) · Scope: 0:00–39:04 (closing show promo excluded)
Hosts: Elad Gil and a co-host (named on audio only as "Elad and I"; labeled Host 2) · Guest: Greg Brockman, co-founder and president of OpenAI, as introduced on audio.
Notes desk: 2026-10-16
Contest OUT · Third-party names (OpenAI, ChatGPT, GPT-4, Math Academy, AlphaFold, Slack, iMessage, XKCD, Heartbleed, Socket, CrowdStrike, Steve Jobs, iMac, etc.) kept verbatim when spoken.
Disclaimer: https://legal.codesolutionsllc.com/si-disclaimer — Notes of a public interview — not Code Solutions LLC views or advice.
Lane: Digest More feed only. Not a Daily X release from this pack alone.
[0:00] Greg Brockman: 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. … The barrier to entry is becoming so cheap. You don't just want software, you want software that is reliable, that is secure, that you can believe in. So cyber spend, security spend should actually vastly outstrip software spend. People have not yet really internalized how much the physics around cyber security are going to change. I think there's an opportunity to end up with a much more secure world.
[0:30] Host 2: 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, [unclear], 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 [unclear]. Here's our conversation with Greg Brockman. (host-said)
[1:00] Host 2: Greg, thanks so much for doing this. Elad Gil: Thanks for joining us. Greg Brockman: Thank you for having me.
[1:05] Host 2: So every week in AI feels like a year. And a lot's happened this week for OpenAI. You rolled out [unclear] to over a billion people [unverified], which is pretty amazing. There's some really exciting news that you have on the math side. … What are you most excited about that's happening right now, both at OpenAI and in the broader AI world? (host-said)
[1:30] Greg Brockman: I have this belief about how OpenAI operates, that there's this cycling quote of, it never gets easier, you just go faster. Every time you think you're at peak impact, you realize there's another level possible. It's not about any one company or technology; it's about we as a world and society coming together to adopt this technology to improve everyone's lives.
[2:00] Greg Brockman: To me, the math results this week represent a real milestone. If you think about the fundamental driver of human quality of life improving over the past thousand years, it's technology. It's scientific advancement. And I think that we now have the ability to massively accelerate scientific discovery. There's always been this question of, is AI actually able to come up with new knowledge? And at this point, it's just impossible to deny. It is definitive [unverified]. There's still more to go; we're not yet coming up with whole grand new theories…
[2:30] Greg Brockman: …but we are able to resolve these [unclear]-year-old math problems. It's not just that the machine gives you an answer; it's what a human equipped with these tools can achieve that was unthinkable before. The leverage of humans is going way up. If you have a math problem, you can solve it. What about applying that same capability to biology —
[3:00] Greg Brockman: — or a small business owner who wants to operate their business well? That level of intelligence tuned for the domain means results aligned with your goals will be within reach of everyone.
[3:10] Elad Gil: The rate of progress has been amazing. Early models had problems with arithmetic; now you're solving advanced theorems. As these advance into math, physics, materials, is there a need to focus on how to communicate that effectively to people?
[4:00] Elad Gil: Only a very small number of people can adjudicate a theorem in any field. Greg Brockman: Yes. Elad Gil: So you have this specialization.
[4:10] Greg Brockman: I came from a math background. I thought I was going to be a mathematician. I was inspired by mathematicians [unclear] working on a [unclear]-year time horizon. There is something deeply beautiful about mathematics being about understanding the structure of the universe.
[4:30] Greg Brockman: There's the question, is math [unclear] or discovered? In my mind the beauty and meaning of mathematics is 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, and it's about simplicity and compression theory, complexity theory … but there's something that is also importantly human about it.
[5:00] Greg Brockman: 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. There's promising signs of life: there's beautiful insights within the work that these AIs have done. We published about [unclear] very important open problems that are now resolved [unverified].
[5:30] Greg Brockman: The mathematical community is going through all of that to look for insights. Domain experts say there are very surprising ways of applying ideas: wait, why did they use that result over here? What's coming out is not these incomprehensible theorems or proofs — which, by the way, there is a history of incomprehensible proofs in mathematics, like the four-color theorem is a perfect example of this. It's like, okay, we just had the machine check a bunch of cases, there's no beauty in that.
[6:00] Greg Brockman: And so I think we're into a different phase of machine discovery.
[6:05] Host 2: I have some optimism that as models also get to better theory of mind, like the ability to figure out what we as humans know and how to get us to follow them on the path to new understanding, will also be paved, right? [seam ~6:10–6:30 filled from audio check] And so, for example, I'm a math person, but at some point in school, I stopped learning math. Then I did Math Academy, then ChatGPT to help me learn new math. The rising tide of what knowledge you can expect people to understand will also be there.
[6:50] Greg Brockman: A really important, maybe underappreciated benefit I expect, hope for and am starting to see from this technology is helping people know more, understand more, understand each other and the world better. There are technical mechanisms: an AI that solves these problems, sees connections between domains — but how does it communicate it to you? There are connections to AI safety and alignment, on having AI that we're able to understand what they're doing.
[7:30] Greg Brockman: 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, 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 really hard to know, is the summary any good, unless you've read the whole book.
[8:00] Greg Brockman: Unless you have another AI trying to prove the summary is bad. If that AI says, here's this paragraph in the book with a very important plot point not represented in this summary —
[8:30] Greg Brockman: — you don't have to read the whole book to see that's a good argument. There are ways to train AI that are very aligned with humans having oversight, being in the driver's seat — a fundamental value for the future from OpenAI's perspective —
[9:00] Greg Brockman: — and aligned with making AI results something you learn from, tuned for the human, rather than abstract knowledge.
[9:05] Host 2: For most people, advancements in math theory are still abstract. And on AI for science, there's a claim it could solve all disease in some number of years. How do you look forward one step at what we might see in between? (host-said)
[9:20] Greg Brockman: We've had years and iterations of this technology to see the shape of what to expect. When an AI is capable in a dimension, and you understand why, then other fields with similar properties should expect similar results.
[9:30] Greg Brockman: Why has mathematics been transformed? Partly our AI is at a certain level of capability and raw intelligence. Partly mathematics is checkable — but that's not quite true anymore. We started with arithmetic, a numeric answer you can check. Proofs are very hard to check.
[10:00] Greg Brockman: A formalized proof is checkable. Natural-language proof, which is what we're producing, is much more difficult. The core insight is really about data. You can get lots of data in mathematics, in theoretical physics. Biology is very data bottlenecked.
[10:30] Greg Brockman: We've been working on techniques to work with less data. The fundamental physics of deep learning: compute, data, algorithms equals capability, equals results. We're able to do it in hard domains where the smartest people spent hundreds of years, and now the AI can solve it [unverified].
[11:00] Greg Brockman: Transferring to biology and drug discovery, the question — not yet resolved — is do we have the data, can we get it? We're already applying these models and techniques in the biological domain. Some areas will fall quickly, a massive acceleration —
[11:30] Greg Brockman: — of things that used to be painful. But there will be other bottlenecks, some not about deep biological theory. Regulatory filings are a huge bottleneck: it takes like a year to write up your case for a clinical trial [unverified]; you can write those with an AI much more easily. We'll look at every step of drug discovery —
[12:00] Greg Brockman: — places where the data doesn't exist, so innovations in wet labs and simulations become very important. Breakthroughs are coming, but some things are shared with mathematics and some are very different in terms of where the data comes from.
[12:10] Elad Gil: You mentioned biology, and I think that's a great example where a lot of the early research, AlphaFold and things like that, were really focused on designing a molecule or understanding how something binds to each other.
[12:30] Elad Gil: And to your point, it takes 15 years to develop a drug. It's $1.5 billion. $15 million or so tends to be pre-clinical, mouse studies, the molecule, and then 1.4 billion plus tends to be the clinical trial [all figures unverified]. And it's also two, three years versus another 12 years. 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.
[13:00] Host 2: How do you at OpenAI stack rank areas of scientific endeavor in terms of focus — generating data, results or capabilities others can benefit from?
[13:10] Greg Brockman: Drug discovery is the most clear and obvious area if you get to pick one to fully revolutionize. You're going to uplift everyone's quality of life; maybe the top one application of AI everyone's been thinking about for a long time.
[13:36] Greg Brockman: There's this whole pipeline; it's so long. We're going to find humans are incredibly bad at biology, at drug discovery. Natural language, humans are well suited to it. Biology is totally foreign.
[14:02] Greg Brockman: For a neural net, there's no reason to believe it should be better at natural language than biology. Given the results in natural language, from first principles you should be able to crack the language of biology. That is number one. We're so close on so many axes. Host 2: Very exciting. Greg Brockman: You can massively increase parallelism and decrease many of these steps in duration.
[14:35] Greg Brockman: There are many dimensions to drug discovery, for example n-of-1 diseases. Lots of people for whom the drug to cure their condition exists out there in the literature, in a paper; find that one paper and you can cure the person. Plenty of precedent. We can massively scale up those operations, so short-run impact is possible. Material science is another example, and a long tail of other domains.
[15:12] Host 2: Part of why a drug might take 15 years to market is successively more expensive trials, safety considerations, large-scale human trials, ordained by regulation, for good reason. (host-said)
[15:34] Host 2: A limiting factor, if the technology works and data can be acquired or generated, is convincing regulators to change expectations around which drugs are accepted and what the steps are.
[15:51] Greg Brockman: There's more work to reach this point, but if you can prove the physics of drug discovery have changed, it should trickle through to every part of the system.
[16:05] Greg Brockman: Take self-driving cars. How roads operate and safety are tuned for human drivers.
[16:18] Greg Brockman: If self-driving cars are ten or a hundred times safer than human ones — AI delivering results so unambiguously what you want — you should say, how do we rethink the whole transportation system?
[16:36] Greg Brockman: Maybe regulation that says only self-driving cars in certain areas. Changes need to trickle back into regulation. Regulation is not uniform: very different regulations in Australia versus the US for clinical trials.
[17:00] Greg Brockman: You're already seeing people moving trials to places where the regulatory environment makes more sense for how they want to pursue them. How things should change will depend on what we can actually prove.
[17:21] Greg Brockman: Do you reach the point where you get better results from a neural net than from running a clinical trial? We're not there. Host (unconfirmed): Yes. Greg Brockman: But you could imagine — Host (unconfirmed): If they were more predictive on efficacy or tox, obviously you should change it.
[17:35] Greg Brockman: Of course. Society should, and I hope will, lean into what is possible having fundamentally changed, and quickly reorient. You have to think through all the consequences and implications.
[17:58] Greg Brockman: For long-term studies, whether a neural net's simulation is accurate for long-term outcomes, it's hard to not have at least one long time tick to correspond to that.
[18:15] Greg Brockman: But I think we will see 80% of what is possible come to fruition very quickly [unverified].
[18:24] Host (unconfirmed): You're a little ways out from the App Store release and DevDay. Anything in how people use the new products that surprises you?
[18:38] Greg Brockman: Dots has really changed how many of us work. Host (unconfirmed): Can you describe Dots briefly? Greg Brockman: Dots is a proactive agent. Always on, always working. You stop working, it keeps working.
[18:53] Greg Brockman: My Dot looks out for work I was supposed to do: messages I was supposed to send, things I said I'd do. I wake up in the morning, there's a list — oh yeah, this is my to-do list.
[19:09] Greg Brockman: It often notices things I never asked it to, like a conflict on my calendar in terms of scheduling, it'll point that out. Or, hey, you actually need to leave this meeting earlier in order to get to the other thing.
[19:22] Greg Brockman: And so it's so contextually aware. It knows what I'm doing with my day. It understands the projects we're working on. [seam ~19:22–19:30 filled from audio check] It knows what's in my inbox, knows what's in my Slack, and is able to integrate across these, and knows the questions I've been asking about, so it knows what areas are most top of mind for me. In some ways, what Dots is, is a change to the form factor, not the capability. It's an [unclear] level model. It's got its own cloud computer. But the change in that form factor in a workplace setting with this level of intelligence, it's a game changer.
[19:55] Elad Gil: Do you have multiple Dots running, or just one covering all your background tasks?
[20:00] Greg Brockman: I've got one. A little owl is the icon. You can actually change the icon. It's just sort of doing everything across my work life.
[20:10] Elad Gil: OpenAI has research efforts and a consumer, enterprise and developer/platform business. Will there be a separation between assistants that help you at work, like Dots, and consumer assistants, like the [unclear] of the world?
[20:30] Greg Brockman: Look, I think these things are going to converge. No question. There are some pressures that are very different in different environments. For a mass-market consumer product, figuring out how to scale and provide things for free or at very low cost is very important. Whereas in the workplace, it is much easier to say, "Peak intelligence, and as long as you're getting returns to paying more, it's worthwhile to throw more compute at it."
[21:00] Greg Brockman: The main distinction is really around cost. We've seen this in ChatGPT: we put in a lot of effort to have the largest-scale free tier, and bringing things like memory to all of our free users was an incredible endeavor, a real huge engineering effort. It's very easy to implement the same functionality at much higher cost for a very small number of people.
[21:30] Greg Brockman: In the personal assistant space there are going to be different market segments where different amounts of compute are going to be expended. Sometimes that means the product can look a little different. And different use cases: payments and travel and things like that are much more for the consumer.
[22:00] Greg Brockman: But I don't think there's any fundamental difference. It's really just an AI that's doing things, and it just happens to have slight pressures one way or another. One of the strengths of OpenAI is the fact that we have 1.2 billion consumers [unverified; figure from audio check] on ChatGPT, and we also have the leading coding and enterprise business, and no one else has both in-house. A lot of our strategy is to unify these so we're not competing on two fronts. It's really competing on one front.
[22:30] Greg Brockman: This is also the general direction of travel of AI, to blur the distinction between consumer and enterprise. We're going to have this renaissance of entrepreneurs. We're already seeing people quitting their job to start companies who say, "I never would have done it if I didn't have an AI I knew I could rely on that could take care of so much of the normal barriers to entry."
[23:00] Greg Brockman: The distinctions that have held for so long, I think they're going to change.
[23:05] Elad Gil: What traditionally separated enterprise and consumer is UI. Besides cost and fidelity, the user interface in an enterprise setting is often very different from consumer.
[23:30] Elad Gil: You're looking toward unifying these layers. You've also been thinking about dynamic UIs — dynamically generating a UI useful for a user — or these things move into the background and it's part of your chat. How do you think UI will evolve for these use cases?
[24:00] Greg Brockman: First of all, I think the best UI is no UI. That is where we're going. You have a thing you can talk to: verbally, over Slack, iMessage, any of the surfaces; you can call it. That's literally what you can do with Dots today. But there are fundamental differences between a personal setting and a workplace setting. In a workplace you have an admin, your company security team.
[24:30] Greg Brockman: You have more stakeholders. It's not just an AI that's for you, representing you, that you own and whose outcomes you own. That's the main distinction I see between enterprise and home. It leads to product differences, sometimes small: when we release a new feature, enterprises want control over when it's introduced, so we have admin controls and toggles.
[25:00] Greg Brockman: So you can have an experience with less functionality available because it's in the workplace. And the whole point was that there's no UI, so what does it mean for functionality to be missing, and how do you surface it and let people know? That's an ongoing challenge; it will not magically go away, and you need to lean into that complexity.
[25:30] Greg Brockman: To me it's kind of crazy and beautiful: ChatGPT started as a text box. What we're heading to, there isn't even a text box. That is the AI we were promised.
[25:45] Elad Gil: What makes a product without a UI better: how much compute I can spend, how good the model is, a memory system, decisions about behavior, how well models are calibrated on how hard they should think about something?
[26:00] Greg Brockman: That's something we're getting very good at. With the release of [unclear] to everyone, if the AI is going to think hard about a problem, it can give you a quick answer, and then keep thinking and then give you the more correct, the more fulsome answer. [seam ~26:00–26:10 filled from audio check] 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. 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 if you think about even something like flight booking, you want kind of a complex UI there, right? You can see all the different options.
[26:25] Greg Brockman: For any question you might ask, like how do I assemble a bicycle, you might want some very bespoke app that a person would have written for a month. You're never going to hire someone to write a one-off app to assemble a bike, but it's fine for an AI to do that in real time. The AI is molding itself much more to you; the user interface is becoming exactly the thing you need at the moment you need it. That is a lot of our focus, across every aspect of the product.
[27:02] Elad Gil: One of the biggest shifts in the discourse around AI has been around cyber — building secure systems resistant to AI-driven approaches. How are you building guardrails and approaching cybersecurity?
[27:15] Greg Brockman: The physics of cybersecurity are changing, and we're in what we call the defender's window. We have peak frontier AI capabilities that can help defenders secure themselves. They're not broadly diffused yet. But broadly diffused capabilities are ratcheting, and you can see this from model release to model release from providers across the world.
[27:37] Greg Brockman: 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. 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 [unverified]. 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 very capable models, to find and fix vulnerabilities.
[28:20] Greg Brockman: These flaws have existed. Every line of human-written code, you have to ask, is this secure? The way we built systems was fundamentally insecure, and we always knew it. There's never been a CISO who's been over-resourced.
[28:40] Host 2: You can look at the XKCD comic of all the world's infrastructure reliant on one open-source project written by one dude in his basement. Or Heartbleed — anything that predated these AI issues.
[28:56] Greg Brockman: Exactly. These are really human issues. The way we chose to build the internet, security got bolted on. Early worms and viruses, then virus-scanning software, encryption, HTTP should be HTTPS. We took the slower route to add security afterwards. The internet's been built for resilience, not necessarily for security.
[29:37] Greg Brockman: And I think that what we're going to see now is a moment where a lot of that tech debt is going to come to the surface. It's time for us to pay it down. As a company, we are really leaning into trying to help with this moment. So for example, we have a [unclear] 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?
[30:16] Greg Brockman: People have not yet internalized how much the physics around cybersecurity are going to change. There's an opportunity to end up with a much more secure world. AI lets you pay down tech debt in a way previously unthinkable. Formal verification of software: these models are great at proving things, so why not prove security properties of software? The technology has been there, but it's just so much work.
[30:45] Host (unconfirmed): Painful, niche, expensive. Yeah. Greg Brockman: Exactly. No one does it for anything significant.
[30:49] Greg Brockman: And so I think all of that can change. 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 [unverified] and said, all of your projects are on hold, you are just securing now. And took our peak intelligence model at the time, had it poke at all of our infrastructure, built some tools so that it's easy to do it again over time. We found a number of issues. We fixed them. At some point we saturated the issues that model could find.
[31:26] Greg Brockman: We'll have a next model and repeat that process. And so much code is now AI-assisted; we are training these models to write secure code from the beginning. We can have fundamentally more secure systems through these AI models.
[31:54] Elad Gil: So there may be a period with more cyber attacks and incidents as people catch up — codebases getting more secure, more AI-generated code pre-set to be secure. A transitionary period, more offense and defense, and then defense will catch up. 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.
[32:22] Greg Brockman: Yeah. Look, I have a lot of optimism about how we can navigate this moment. I think it does require a coordinated response, right? [seam ~32:22–32:30 filled from audio check] It really requires defenders to internalize how the physics of cybersecurity are changing, that there is this window, and that window is not going to last forever. The security of an organization will relate to the fundamentals they've gotten right, a one-time investment, and the level of cyber capability they use to defend themselves.
[33:00] Greg Brockman: In that world, defenders can outstrip attackers. But it takes change in how every organization operates, how we think about critical infrastructure, and resourcing those who haven't had those resources.
[33:21] Host 2: Your description is like internet-wide tech debt, and your internal response is a really good one. I've been part of companies where tech debt is painful to address — taking 25% of the engineers, or a feature freeze for a quarter…
[33:45] Host 2: That doesn't feel like the state the world has decided on. Security investment is traditionally low single-digit percent of the IT budget [host-said]. What should it be now? Is traditional network, endpoint, corporate security less relevant, or do we just spend twice as much?
[34:16] Greg Brockman: I have a feeling that cyber spend, security spend should actually vastly outstrip software spend. Host 2: Well, that's a statement. Greg Brockman: I really believe it. I may not be right, but software, the barrier to entry… Host 2: It's free-ish.
[34:38] Greg Brockman: It's becoming so cheap. You don't just want software, you want software that is reliable, secure, that you can believe in. The resilience of that software… Host 2: That's the value. Greg Brockman: Exactly, that's going to become even more important.
[35:00] Host 2: You pointed to the xkcd with the little block at the bottom of some open-source maintainer. That never made any sense.
[35:10] Greg Brockman: We had that world because of how hard it is to write reliable software. Everyone has a supply chain and so few companies audit it. Someone uploads a malicious package and a number of companies get hit. All of that needs to change…
[35:30] Greg Brockman: …and it's impossible to change with human developers alone, because so few people have this obscure expertise in packages and security techniques.
[35:38] Elad Gil: Third-party software like Socket will scan your open-source supply chain for vulnerabilities. And there's the human element, maintainers taking over a project and installing something malicious.
[36:00] Elad Gil: It's 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 this market.
[36:06] Greg Brockman: Exactly. 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, because we have the frontier model, we have the peak intelligence model, certainly in cybersecurity [unverified]. We think that there is transformative potential for good possible.
[36:30] Greg Brockman: That means that we as a company have been going through a security transition. Over the past couple of months, we've talked about safety, security, alignment of our development process, and also how to help the world become more fundamentally secure through this period.
[36:44] Elad Gil: OpenAI has always been very pro-human, pro human agency. The original premise was to counterbalance other companies, creating AI used broadly in a positive way. Has that ethos continued?
[37:05] Greg Brockman: That ethos is core to our mission, to ensure AGI benefits all of humanity. Humans should be firmly in the driver's seat, have control over the future, and this technology should amplify human agency. Host 2: Even as models surpass humans in capability.
[37:21] Greg Brockman: And I think this is absolutely critical. Keep in mind these models are very jagged. There are places where obviously these AIs are more capable than anyone today. Even GPT-4 [unclear] far more than any person.
[37:37] Greg Brockman: And I'm reminded, for example, of the story of why Steve Jobs put the handle on the iMac, so that people could feel that they could control the machine; you can always pick it up and throw it out the window. The same spirit needs to apply to AI.
[38:00] Greg Brockman: Humans in the driver's seat: human judgment, human values. Maybe humans in certain fields will operate more like CEOs or board members. It's important that we have that level of agency.
[38:30] Greg Brockman: As a CEO you have domain experts, teams with deep knowledge, maybe better ideas than you on many axes. But the question is: are you happy, is this what you wanted, does this make your life better? It's not just about AI, it's about technology. Technology is to lift up human civilization — the single most fundamental driver of improvement of quality of life — and that is what we should be seeking from AI. That is the belief of OpenAI.
[38:59] Elad Gil: Thank you so much for doing this, Greg. Greg Brockman: Yeah.
[39:00] Elad Gil: Yeah, thanks so much for your time.
[39:01] Greg Brockman: Thank you both. Appreciate it.
[39:04–39:22 show promo — not transcribed.]