[BidClub_]
Moonshots · · 82 min

Open AI Insider on GPT-5, AGI & the Great AI Race w/ Kevin Weil & Dave Blundin

Peter DiamandisKevin WeilDavid Blundin

YouTube ↗
TL;DR

  • GPT-5 was presented as OpenAI’s effort to combine frontier capabilities into one broadly useful product, while the interview left AGI unresolved. Kevin Weil called it OpenAI’s smartest model and emphasized coding, health, complex instruction-following, tool use, and agentic work; pricing came in at “less than half” the prior level. Yet OpenAI still cannot reliably predict what a model will unlock: capabilities appear “through the mist,” sometimes only after release.

  • Compute, not customer demand, is OpenAI’s binding constraint. Weil said the company uses essentially the same model settings as customers, remains “completely maxed out at all times,” and finds an immediate use for every new GPU—lower latency, faster tokens, wider product access, or more research experiments. Stargate’s more than $500 billion of planned infrastructure is therefore a capacity expansion into what he called “basically infinite demand for GPUs within these walls.”

  • OpenAI’s distribution strategy reverses conventional software monetization: expensive capabilities begin in paid tiers, then migrate toward free. Deep Research moved from Pro to Plus and eventually limited free access; India received a heavily discounted plan with roughly 10x the usage of a free account. Diamandis said models are now about 100x cheaper than GPT-4 was at launch even as intelligence increased, while Weil said paid tiers will retain the most computationally intensive work.

  • The startup test is whether better foundation models strengthen the product or erase it. Weil advised founders to build where current models show “little glimmers of hope,” so the next release makes the application “sing,” rather than patching a limitation that OpenAI may soon remove. His premise is sweeping: practically every scaled product, service, and device predates AI and “they’re all going to be reinvented.”

  • OpenAI’s envisioned AGI product is ambient, proactive, and capable of generating disposable software—not merely a smarter chat window. Weil expects interfaces to be created in real time, routine work to be completed before the user asks, and an assistant that can “see what you see” and keep “chugging away behind the scenes.” The Jony Ive question remained unanswered when the interview was cut.

  • Reasoning adds a second scaling axis beyond pretraining: how long a model can work while staying on track. ChatGPT reasoning may run for roughly 60 seconds and Deep Research for 20–30 minutes, but Weil sees no reason models could not work for days, months, or years; “the longer the models think, the smarter they get.” Well-specified problems such as chip layout can then turn compute into iterative gains because every candidate design can be scored.

  • AGI may arrive as a rising capability gradient rather than a clean threshold event. Weil argued that today’s models already outperform him in some domains and remain clearly inferior in others, while “the level of water is rising”; the hosts compared that transition with society barely noticing that AI had passed the Turing test. The measurement challenge is shifting from saturated, easily graded benchmarks toward ambiguous but economically valuable work.

  • The human hedge in an increasingly automated economy is purpose, personal connection, and deployment into consequential institutions. Weil rejected futures where people merely “eat grapes and write poetry and receive our UBI,” arguing that saved time gets redirected toward larger goals and that in-person connection will matter more. His parallel role as an Army lieutenant colonel reflects the episode’s broader deployment thesis: superior models provide little advantage “if they’re sitting on the shelf” while rivals integrate weaker systems everywhere.

Digest · the substance, structured for research

1. GPT-5 trades spectacle for breadth, price, and usability

  • Weil described GPT-5 as probably “the most anticipated AI launch of all time,” but framed the substance as breadth: OpenAI’s smartest model, a strong coding system, and a model capable of complex instruction chains, numerous tool calls, service integration, and agentic execution “without losing the plot.”

  • Health received unusual product attention because users already bring ChatGPT everything from a child’s new symptom to cancer diagnoses and associated data. Weil’s boundary was explicit: “It doesn’t replace a doctor,” but a capable system available for conversation 24/7 can still help people think through information.

  • Diamandis contrasted expectations of an AGI reveal with the delivered product: a more useful unified system at less than half the prior price point. The hosts likened its one-product simplicity to the original Google search box—an important interface advantage for technology whose possible uses are otherwise difficult to explain.

  • OpenAI itself does not know the complete capability envelope in advance. Researchers may see a model “coming through the mist,” but Weil said emergent properties can surprise the team during development or even become visible only after hundreds of millions of users begin testing it in unanticipated ways.

2. Seven hundred million users turned launch day into the real evaluation

  • Internal and external testing cannot reproduce what happens when a model reaches 700 million people. OpenAI therefore combines feedback from Reddit, Twitter, LinkedIn, customer support, friends, and usage data; that data showed one of the company’s biggest waves of Plus upgrades around GPT-5.

  • Commercial strength did not negate product misses. Weil conceded that GPT-5’s initial personality felt “a little bit wooden,” and the team rapidly shipped a warmer version—an example of deployment operating as continued product development rather than a terminal release.

  • The hosts argued that voice has crossed a qualitative threshold: instead of abandoning a conversation after a minute or two, a user can stay engaged for hours on a drive. Weil’s children treat a 20-minute voice conversation with ChatGPT as entirely ordinary, illustrating how quickly interface expectations can reset.

  • The broader uncertainty is not limited to personality. Weil rejected the idea that anyone at OpenAI knows exactly what will exist three months or a year ahead: research ideas can begin working, but what those capabilities enable becomes clear only as the model and product come together.

3. Iterative deployment makes free access the destination

  • Weil rejected the premise that OpenAI meters out mature capabilities on a secret schedule. Its stated mission leads toward “iterative development, iterative deployment”: release as soon as a system is ready, subject to safety checks, then let users build, expose shortcomings, and feed those lessons into the next model.

  • Internal systems are ahead of public products, but Weil characterized them as research and development—not finished intelligence being strategically withheld. OpenAI would rather release early than wait to bestow something “fully formed upon the world,” because external experimentation is itself part of gaining mastery over a capability.

  • The monetization path runs opposite to conventional software. Features begin in Pro or the $20-per-month Plus tier because they are expensive; OpenAI then works to move them outward. Deep Research began in Pro, reached Plus, and eventually gave free users a monthly allowance.

  • Free distribution and premium economics can coexist, in Weil’s view. OpenAI will maximize the useful intelligence available without charge, while increasingly valuable, compute-heavy tasks remain paid—especially as ChatGPT moves from a reactive system awaiting prompts toward one that works proactively while the user sleeps.

4. India is the clearest test of intelligence becoming mass infrastructure

  • India matters to OpenAI not merely as a large market but as a youthful population with extensive latent building capacity. Weil described strong engagement among developers and API customers, while Diamandis emphasized the country’s 1.4 billion people and potential applications across education, healthcare, and governance.

  • Weil’s software thesis begins with roughly 30 million people worldwide who can currently code. AI coding models could expand effective software creation to 300 million and eventually three billion people; each order-of-magnitude increase in access to such a general-purpose capability, he argued, changes the world.

  • On the morning of the interview, OpenAI launched an India-only paid tier priced far below its standard plan and offering about 10x the access of a free account. The discount expresses the access mission, but Weil kept the constraint visible: GPUs cost money and remain limited.

  • For countries asking about “sovereign AI,” Weil returned to falling access costs: ChatGPT can be used free on a phone without signing in, while Diamandis said current intelligence is roughly 100x cheaper than GPT-4 was at launch. Weil endorsed the broader cost curve rather than detailing a separate sovereign architecture.

5. Competition accelerates OpenAI, but focus is its claimed moat

  • Weil readily acknowledged Google as a fast-moving competitor building good models, alongside Anthropic and “a bunch of other players.” Competition benefits consumers and businesses and motivates OpenAI to move faster; as he summarized a friend’s maxim, “capitalism is undefeated.”

  • OpenAI watches competing products because “there are smart people within these walls and there’s a lot more smart people outside of these walls.” Rival features sit alongside customer requests and observed use as inputs, but Weil said the decisive variable remains internal execution against a clear AGI mission.

  • Diamandis raised Google’s roughly $100 billion in free cash flow, long infrastructure history, and massive data centers as a structural vulnerability for OpenAI. Weil’s rebuttal was concentration: Google supports many products, while OpenAI has “one product” and “one mission”; building AGI is existential rather than one initiative among many.

6. Founders should build for the next model, not defend against it

  • Weil’s opportunity map starts with a reset: virtually every product, service, and device operating at scale was designed before AI. Adding AI is analogous to adding electricity to mechanical systems, and history suggests many incumbents will fail to perform the reinvention themselves.

  • OpenAI will participate in that rebuilding, but cannot cover medicine, material science, technology, and every other industry. That leaves an enormous startup surface even if the model provider expands horizontally.

  • The best position is at the model frontier, where an application only barely works and founders can see “little glimmers of hope.” If a new model arrives two months later and makes that fragile capability “sing,” the application was aligned with progress rather than exposed to it.

  • The dangerous position is a wrapper whose value consists of repairing today’s model deficiency. Weil’s blunt test: if the next OpenAI launch could “obviate the need for the thing you’re building,” do not build it. Diamandis connected this to Perplexity founder Aravind Srinivas’s phrase “AI complete”: ride the capability wave instead of being swamped by it.

7. The AGI interface becomes generated, ambient, and potentially neural

  • Chat remains powerful because it approximates the generality of human communication—writing, speaking, and exchanging visual context—but Weil does not see it as the only AGI interface. A sufficiently capable model should generate the most economical UI for each task in real time.

  • The discussion extended that vision to ephemeral software: purpose-built tools can be created for an immediate need and discarded because recreating them takes an instant. Real-time image and video generation could also visualize designs, construct scenes, or produce entertainment around the user.

  • Weil’s preferred metaphor is the “jewel in your ear” from Ender’s Game: an intelligence that can see what the user sees, hear what the user hears, and access broad information. ChatGPT should not wait inside an app; with the context the user gives it, it should keep “constantly chugging away behind the scenes,” acting and saving time.

  • BCI produced a real disagreement. Weil initially argued that AI already moves too quickly for additional input bandwidth to matter, while Diamandis argued that visual and neural access could still help. Weil later said he would personally adopt a BCI, and both qualified that interest by the question of safety. The unresolved question is whether neural access expands cognition or merely accelerates an interface humans already struggle to absorb.

8. Automation raises the value of shared culture, physical presence, and purpose

  • AI-generated media likely becomes hyperpersonalized, with films made for a single viewer. Weil preserved the tradeoff: individualized content is compelling, but society loses the shared experience of everyone seeing the same news broadcast or theatrical release; at the opposite extreme, explicitly human-made work may become more valuable.

  • The hosts proposed that the virtual world could change at “warp speed” while physical infrastructure changes more slowly. Weil’s invariant was human nature: face-to-face work, handwritten notes, and personal connection still matter, and may matter more when digital production becomes abundant.

  • Purpose is the dividing line between Weil’s “Star Trek universe” and “Mad Max universe.” He rejected the image of people sitting back to “eat grapes and write poetry and receive our UBI”; previous labor-saving tools did not eliminate ambition, and AI should redirect saved time toward more interesting work.

  • Every one of eight billion people still receives seven days a week, 24 hours a day, and 365 days a year. Diamandis called AI the greatest time multiplier; Weil’s corollary was that the human drive to pursue something larger and leave the world better “is not going to change”—AI can only supercharge it.

9. OpenAI’s operating advantage is a tight research-to-product loop

  • Weil estimated the research organization only loosely as “a few hundred.” Its work spans highly academic experiments that may fail or remain unseen for years, post-training tied closely to customer needs, and every point between those poles.

  • OpenAI’s distinctive mechanism, in his telling, is the loop among research, engineering, and product: create a capability, turn it into a product, collect feedback, and return that evidence to model development. Deep Research and agent functionality emerged from this pattern.

  • Responsibility is divided rather than concentrated in the CPO role: Weil oversees products such as ChatGPT and the API; Mark leads research; Greg leads data-center infrastructure and scaling. He also emphasized co-location and whiteboard work in what remains “a very research-oriented culture.”

  • A physicist by background, Weil had nearly joined a fusion company after Sam introduced him to companies in the field. He later contrasted his previous employers with OpenAI: Facebook and Instagram moved quickly in his experience, but “nothing in my experience compares to OpenAI.”

10. Saturated benchmarks push AGI measurement toward messy economic work

  • Traditional AI benchmarks are now “almost all saturated,” forcing researchers toward harder tests such as FrontierMath and ARC-AGI. Saturation demonstrates model progress, but also removes the simple yardsticks that made improvement legible.

  • OpenAI increasingly evaluates tasks connected to economic value: interpreting health cases, constructing a company financial model, or performing work associated with doctors, lawyers, and bankers. These resemble one definition of AGI—performing economically valuable tasks—but lack a single canonical answer.

  • The grading problem grows as capability broadens. Mathematics is comparatively easy to score; financial models can be built several valid ways, and creative writing has no uniquely correct output. The models are entering precisely those useful domains where evaluation becomes “softer” and self-improvement harder to supervise.

  • The hosts observed that AI passed the Turing test with little ceremony after it had stood for 50 or 60 years as a landmark. Weil expects AGI to feel similarly gradual: models are already far smarter than him in some areas and clearly weaker in others, but “the level of water is rising.” Diamandis cited an IQ measurement near 148 for GPT-5 Pro; he also said GPT-5 had won an IMO gold medal, while Weil noted second place in one programming competition.

11. Messy model names are a side effect of shipping capabilities early

  • Weil accepted “well-deserved flak” for a catalog that simultaneously contained o4-mini and 4o-mini. The confusion reflects iterative deployment: OpenAI prefers releasing a specialized breakthrough quickly rather than waiting until every capability fits an elegant universal product.

  • GPT-4o introduced broader interaction, including speech; the separate o1 line introduced reasoning—trying hypotheses, rejecting failures, and continuing rather than answering immediately. Early reasoning models excelled at hard scientific problems but were not necessarily the right choice for relationship advice or routine factual questions.

  • Specialization allowed faster observation of where reasoning worked, what users attempted, and where it failed. GPT-5 became the point at which OpenAI recombined those strands into one experience, but Weil expects future experimental offshoots followed by reintegration once the company understands them.

  • Reasoning also creates a second axis of intelligence scaling. Pretraining remains one axis; test-time compute asks how long a model may think while remaining coherent. With o1, o3, and ChatGPT, models may think for about 60 seconds, while Deep Research can spend 20–30 minutes gathering, identifying gaps, and returning for more evidence.

12. Closed-loop chip design can convert more compute into more compute

  • Weil sees no stated reason models could not work for two days, two months, or two years. Andrew Wiles did not solve Fermat’s Last Theorem in five minutes—he worked for seven years—and OpenAI continues to see evidence that longer reasoning enables harder solutions.

  • Chip design is especially attractive because it is constrained and gradeable. A system can propose a layout, simulate it, score the chip’s speed, and iterate; that creates the evaluation loop that open-ended creative domains lack.

  • The hosts argued that AI can also translate algorithmic intent into hardware, narrowing the organizational distance between model and chip teams. Weil said OpenAI is designing its own chips, using AI to improve design and layout, and working with manufacturing partners.

  • His most aggressive conditional claim concerned well-specified problems: apply arbitrary compute to repeated scored iterations and, “so far,” expect arbitrary improvement. He still considered the field fertile for technically strong startups, with material science another underappreciated domain for the same mechanism.

13. Education should use AI to raise the assignment, then aim at frontiers

  • Diamandis rejected school bans on AI because students will use it everywhere after graduation. His proposed redesign was to stop assigning eighth-grade work that AI makes meaningless and instead ask an eighth grader to tackle graduate-level challenges, such as designing a starship with AI support.

  • Weil agreed: assume ChatGPT exists, teach students to use it, take them deeper than a conventional classroom could, and raise expectations for the final work. He cited Ethan Mollick’s Co-Intelligence and Mollick’s practice of redesigning classes around universal AI use rather than pretending it is absent.

  • Even if progress froze at GPT-5, Weil believes existing capabilities would transform society over the next decade; progress, however, is unlikely to freeze. Diamandis extended that premise toward grand challenges in space, energy, and science, arguing that widespread tools let people pursue missions once reserved for kings, queens, and robber barons.

  • Diamandis offered his own conditional frontier forecasts: recurring Starship trips to the Moon by the end of 2026, “definitely” 2027; robot boots on Mars around 2030; Helion targeting 2028 and CFS around 2030. Weil’s broader point was that competition among dense clusters of startups and AI-empowered founders raises the probability that such ambitious programs succeed.

14. In defense and startups alike, deployment beats intelligence on the shelf

  • Weil said he was inducted into the Army as a lieutenant colonel on June 13 alongside Shyam Sankar of Palantir, Meta CTO Boz, and former OpenAI research head Bob McGrew. The program is designed to bring technology and the Department of Defense closer by combining industry expertise with institutional knowledge.

  • Diamandis argued that wearing the uniform makes the group part of the institution rather than outside consultants. Weil agreed that being inside should make them more effective; their work is more ad hoc than a conventional reserve schedule, uses need-to-know compartmentation, and divides focus areas to manage conflicts. Weil expects to concentrate partly on AI for monitoring and improving physical performance.

  • The strategic concern is integration speed: “It does us no good to have the best models in the world if they’re sitting on the shelf” while the PLA uses inferior models but integrates them everywhere. The comment echoed the episode’s broader iterative-deployment logic—usable systems and feedback loops matter more than latent capability.

  • Weil’s final advice to entrepreneurs was to “lean into the AI in every way possible” and assume the steep curve continues. Imagine what should exist but cannot yet be built, then work toward the capabilities likely to arrive in six months, one year, or two; in his categorical formulation, founders who build for that future will benefit because “the models are going to get there.”

Peter Diamandis

You're in the middle of the fastest-growing and most rapidly changing thing in human history.

Kevin Weil

We don't always even know what the next model is going to be great at.

Peter Diamandis

Kevin Weil, the chief product officer, leads all the product functions at OpenAI. Coming off a storied career at companies like Twitter and Facebook, I think everybody knows you.

Kevin Weil

For those who don't know me, I'm Kevin Weil.

Peter Diamandis

I do have an important last question. Has OpenAI fully reached AGI, and what is Jony Ive delivering for you?

Kevin Weil

Oh, man, I've been waiting to talk about this since we started.

Peter Diamandis

Now that's a moonshot, ladies and gentlemen.

Peter Diamandis

Everybody, welcome to Moonshots. I'm here with my moonshot mate David Blundin, and we're here at OpenAI headquarters in San Francisco. We're about to have a deep session with Kevin Weil, the chief product officer of OpenAI. So, let's dive in. Welcome to Moonshots.

Peter Diamandis

Kevin, congrats on GPT-5. That is a huge milestone. As chief product officer, I can't think of a bigger product to be announcing to the world.

Kevin Weil

It was probably the most anticipated AI launch of all time—the most anticipated product launch. We've been talking about it for a while, but we're super excited. We put a lot into it. A lot of the things that people might not even think about—we did a lot of work on the health side.

We added a lot of health data because one of the things we see is that people are using ChatGPT all the time to ask about health data. It could be anything from, “This thing happened to my kid. What should I do?” all the way to, “I just got a cancer diagnosis. Here's a bunch of data. How should I think about this?” It doesn't replace a doctor, but it sure is helpful to have this super-smart thing that you can talk to 24/7.

Peter Diamandis

I love that part of the launch, by the way. Everyone on Twitter has different opinions about that part, but I love that part of the launch. Were you a big part of scripting that whole day, or how did that work?

Kevin Weil

No. We have an incredible team that puts that together. The GPT-5 launch was fun because we had so much to talk about. It wasn't just health. It's also the smartest model we've ever launched. It is an incredible coding model, and it can do a lot of highly agentic things that you may not see as much in ChatGPT but are super valuable for developers building on it.

It can follow very complex instructions, make lots of different tool calls to different services, and integrate things without losing the plot. That has been a real area where the whole industry is trying to make models better. GPT-5 is the best across the board in a whole bunch of areas, and we wanted to figure out how to show all of these things. We needed a lot longer than our normal 20-minute livestreams.

Peter Diamandis

The simplicity of one product serving all was brilliant.

Kevin Weil

There's something cool about models that way, right?

Peter Diamandis

It's like the Google homepage when it was a simple search bar. That's what allowed it to dominate over Yahoo and everybody else. That simplicity is amazing.

AI is so open-ended. You can do anything on that stage. Trying to script that and make it relevant for this really wide audience has got to be extra challenging for the team here to figure out how to do.

Kevin Weil

It's the fun thing about working at OpenAI, too, because we don't always even know what the next model is going to be great at. You have some sense—you can kind of see it coming through the mist at you a little bit—but other times there's something emergent, and we're surprised.

Peter Diamandis

That's the most interesting thing: the fact that AI's properties and capabilities are emergent. No one predicted any of this 2 years ago, let alone 1 year ago.

So, you had to have had—particularly with GPT-5—all these comments coming from the world.

Kevin Weil

Yeah.

Peter Diamandis

People are super excited, and people are critical. How do you deal with all the feedback, and does that feedback actually tie back into how you iterate?

Kevin Weil

Oh, it totally does. These models are so powerful and generic. There are lots of things that we obviously test internally all over the place, and we're testing externally in various ways before we launch. But then you give it to 700 million people, and all of a sudden they're testing it in all kinds of new ways.

We listen a lot. People have feedback on Reddit, Twitter, LinkedIn—all the ways that you'd expect. We have people coming into customer support, and we have friends who are giving us their feedback. You're also just looking at the data, right? The data tells a really good story. We had one of our biggest Plus upgrades.

Peter Diamandis

Oh, my God, I saw those numbers. Stellar.

Kevin Weil

We heard the feedback, too. There were a couple of things that we didn't quite get right. One was that the model's personality was a little bit wooden, so we iterated and shipped a fix very quickly—just a little bit warmer of a personality.

Peter Diamandis

That is an interesting subject in itself, because there's probably not one personality that will meet everyone's needs. It needs to be fairly dynamic to the individual.

I've got to say, the voice models and the interactivity of voice on OpenAI are my favorite of all the large language models. It really feels natural.

Kevin Weil

Oh, really? It crossed a tipping point, too, where if you're on a long drive, you can talk to it for hours now. You used to talk to it for maybe a minute or 2, and then you'd kind of flip out and move to something else. Now you just go for infinite time. It's engaging and interesting.

To my kids, it's completely natural, by the way, that they would take my phone and just talk to ChatGPT for 20 minutes.

Peter Diamandis

You're a new dad as well. How old is your newborn?

Kevin Weil

No, we have an 11-year-old, and then we have 8-year-old twins. We're a little bit into it. We're kind of in the middle stage.

Peter Diamandis

I was looking at the photo of your baby on your WhatsApp.

Kevin Weil

Yeah, that's 10 years old—my WhatsApp photo.

Peter Diamandis

Okay. I mean, you've got to admit, people were expecting AGI.

Kevin Weil

People were expecting GPT-5 to be AGI.

Peter Diamandis

Yeah, I know. I know you will.

Kevin Weil

Yeah. I'm very confident.

Peter Diamandis

Yeah, I'm sure you will. People were expecting that, and then they got a very usable product at less than half the price point.

Kevin Weil

Less than half the price point. Also, what you said a second ago—I really want everybody to understand this: nobody knows what next year's capabilities, or even 3 months from now's capabilities, will be.

A lot of people think, “Hey, no one knows exactly what's going to happen next.” But it's not true that humanity knows what capabilities will and won't exist, because as you scale, it's not obvious.

Peter Diamandis

Are you thinking about consciousness, or what?

Kevin Weil

No. Well, no, I wasn't thinking about consciousness. OpenAI has a bunch of incredible research ideas, and we have some sense of things that are starting to work. We have a rough sense of the capabilities, but what they will enable only becomes clear as the model comes together—and sometimes not even then.

To our earlier discussion, sometimes you launch and then you realize the model has these emergent capabilities that you didn't expect.

David Blundin

When we were here before, I was asking you—because I teach a class at MIT on the foundations of AI ventures, on how to build an AI company—all these people want to be at OpenAI because they want to be in the middle of seeing it happen day to day.

I said, “Why don't you open an office in Boston, where all this talent is?” You were like, “At the rate that AGI will arrive, it won't matter. We're not doing that because it won't matter. The timeline doesn't line up. We're going to keep the headcount small. The AI is going to be the workforce.”

By the time we got a building built and got it populated, it wouldn't matter anyway, because AGI would be here. I don't know if that timeline lines up with what you're seeing, because you're on the inside. You're in the middle of the fastest-growing and most rapidly changing thing in human history.

Kevin Weil

We're very optimistic. There's also something magical about being in person, to your point about a Boston office. You're in person, drawing on whiteboards together. It's a very research-oriented culture.

I keep telling my team that I want the team to come back together again physically.

I want to collocate as much as possible. Zoom is fantastic, but there's human nature.

Peter Diamandis

So, you're constantly creating increasing capabilities within the OpenAI ecosystem, and you've got to decide how much of that capability to launch to the public and how much to hold back. How do you think about that? Because I'm sure you have much more capable models and capabilities than are accessible right now. Is that something that's internally constantly being metered out at a specific rate? How do you think about that?

Kevin Weil

Not really. Our mission is to ensure that AGI benefits all of humanity, and the way that we do that is to put AGI in people's hands as much as possible. We believe in this process of iterative development and iterative deployment. Rather than holding it back and then bestowing it, fully formed, upon the world, we want to get AI out to people as soon as we feel like it's ready.

Peter Diamandis

Yeah.

Kevin Weil

And we'll do that early and we'll do it often. We want to do it safely, obviously, so that's its own sort of guard. But other than that, we'd rather put something out to people earlier and let them play with it, build awesome stuff with it, learn from it, and then get their feedback. That helps us build better models.

It's not the kind of thing where we're like, "There's obviously a bunch of stuff we have internally that's ahead of what we've launched," but that's also stuff that's very much in research and development and will make its way into models over time. It's sort of the ability to learn to harness it. But we try to put stuff in people's hands because the fun thing about this world right now, I think, is that every service, product, and device we use is going to be reinvented.

Peter Diamandis

So that answers the question in terms of core model capabilities, but what about just raw speed? I write code with GPT-5 every single day.

Kevin Weil

Yeah.

Peter Diamandis

And you watch the lines of code come out, and it's incredible. I mean, as a force multiplier, it's like 1,000x more productive than I would be without it.

Kevin Weil

Yeah.

Peter Diamandis

But I'm watching the lines of code come out, and I'm picturing that when you guys are using it internally, it must be many more GPUs blazing fast, with lines of code just pouring out of it. Am I right?

Kevin Weil

Well, we use pretty much the same settings that everybody else does.

Peter Diamandis

You're kidding.

Kevin Weil

But that's really just a function of the number of GPUs we have.

Peter Diamandis

I heard you have a few.

Kevin Weil

We do have a few. We have a little bit more than a few, but we're also completely maxed out at all times. This is one of the reasons that we talk about Project Stargate, where we're going in with a bunch of other groups and building out more than $500 billion of infrastructure.

Peter Diamandis

It's extraordinary. Computronium covering the planet.

Kevin Weil

Totally. I mean, the more GPUs we get, the more they immediately get used, whether we take them on the product side and use them to lower latency or speed up token generation, launch new products, take a product that's only available to Pro users and bring it to Plus users or free users, or simply run more experiments on the research side. There's basically infinite demand for GPUs within these walls, and that's why we're doing so much to build capacity.

Peter Diamandis

I'm really glad you said that, because there's a school of thought out there that GPUs will commoditize, and it's just so wrong. It's so incredibly wrong.

Kevin Weil

We're far from that moment, at the very least.

Peter Diamandis

I think that will never exist in the world.

Kevin Weil

It's one of those things—it's like the internet. Every bit that we lower latency or increase bandwidth on the internet, people do more things. Video used to be impossible; now video is every day because the capabilities are there and the network can handle it. The more GPUs we get, the more AI we'll all use.

Peter Diamandis

Last year, I made 4 trips to India. You know, there's a population of 1.4 billion and a huge number of builders there—a massive latent talent for building. I saw Sam's tweet about building for India, and I'm curious: How do you think about that? India needs AI for education, healthcare, and governance. It needs it to help uplift 1.4 billion people. It's the only thing that's going to do that. How does it enter your mission as chief product officer? Do you think about that differently from just in general?

Kevin Weil

It's a huge priority. I was out there with Sam and a bunch of people from OpenAI 4 or 5 months ago, and the reception was just incredible everywhere we went. We were talking to developers and people using ChatGPT and building on top of our APIs. It's the biggest country in the world, and there's so much we can do. It's also the most youthful.

One of the things I'm really excited about, talking about coding, is that today there are something like 30 million developers in the world—30 million people who know how to code. With an AI coding model, we can give a couple more orders of magnitude of people the ability to code. We can take it from 30 million to 300 million to 3 billion people who effectively know how to code and can create software.

When you can create software, it's such a general-purpose skill. You can build personalized tools and all kinds of things. Every time you increase access to a general-purpose tool by an order of magnitude, the world changes. So whether you're talking about health or education or, frankly, the ability to build software, I think this is why AI is going to be the biggest transformational force in our lives. And I think in places like India, it's going to be that much more powerful.

Peter Diamandis

Yeah. There are nations that need it more than others, and India is most definitely going to be one of the biggest beneficiaries.

Kevin Weil

Yeah, we actually just launched this morning a paid plan in India just for India. It's a much cheaper plan, but it gives Indian users a lot of access to ChatGPT—like 10x more access than you get as a free user.

Peter Diamandis

Nice.

Kevin Weil

It's a heavily discounted plan because we're trying to bring as much of the benefit of AI to people in India as we can.

Peter Diamandis

So how much do you think that's going to unlock a latent talent base? You start in India, but you think globally—there's talent everywhere.

Kevin Weil

Yeah.

Peter Diamandis

And a lot of it is latent. It can't get into a great school. It can't get into a great ecosystem. AI is going to be an incredible leveler of that, starting with coding. Is that part of the motivation for going in with a lower price point in India?

Kevin Weil

Yeah. There's talent everywhere, and there's also hustle and urgency. So, yeah, we want to bring as much AI as we can, and we've got limits on GPUs and all of this. We have to pay for it one way or another, but we want to put this in the hands of as many people as we can.

One of the interesting things about OpenAI, one of the things that's different about this company relative to anywhere I've ever worked, is that normally, when you build products, you move things behind the paywall. A product starts out free, and then at some point you make it a paid feature to try to get people to pay more money for whatever you're building.

We go the opposite way at OpenAI. Stuff starts in Pro or Plus, so you're paying $20 a month for it. Our entire goal is to make that a free product because we want to give a more powerful product to more people, because we think it makes the world a better place.

If you look at the progression of a lot of our things, Deep Research started out as a Pro feature, then we brought it to Plus, and now, as a free user, you can do a bunch of Deep Research every month. Our goal is to make the most powerful product we can and give it to as many people as we can.

Peter Diamandis

So you're probably in charge of balancing the need to create as many GPUs as possible—Stargate—and the fact that you need a massive amount of infrastructure, which takes money.

Mhm. And then you want to balance that against, yeah, but we also want to put this in the hands of as many people as possible, as cheaply as possible. That's got to be incredible tension inside the building, trying to figure out how you balance those 2 objectives.

Kevin Weil

Not so much tension, actually. You might be surprised, because we have a long-term belief. I have a long-term belief that the idea that we used to live our lives without AI helping us every day is going to feel crazy in a few years. And so, if we're building a product that adds a ton of value to people's lives, we're going to make as much as we possibly can available for free, but there are always going to be things that are super computationally heavy and expensive for us that will be in a Plus plan or in a Pro plan.

And if we do our job—if I do my job and the team continues to do their job—then those features are going to be super valuable. They're going to be ever more valuable, right? AI is only going to be able to do more for you while you're sleeping, rather than waiting. ChatGPT today is a pretty reactive product. You go there, you ask a question. It can do things for you that would have been crazy 2 years ago, but you're still asking it first.

Peter Diamandis

You're still starting to prompt.

Kevin Weil

But you can definitely imagine a world where ChatGPT is much more proactive because of what it knows about you.

Peter Diamandis

You prompt it to be proactive. It's actually proactive.

Kevin Weil

Totally. But why not think 24 hours a day? You're asleep, and it's thinking about what it can do for you, you know? So there's always more that we can do.

Peter Diamandis

My favorite model is still Jarvis.

Kevin Weil

I want Jarvis in my life. It's doing things for me all the time. I still want my AI to do surprise and delight. The doorbell rings and something shows up that I'm not expecting, and it says, “I bought this for you because I think you'd like it. I heard the conversation you had.”

I'll do that for you. I'm going to ask ChatGPT, “What random thing should I buy you?” and it will show up at your doorstep. We'll see what it does. That's a Peter Diamandis edition.

Peter Diamandis

Fantastic. I love it.

Kevin Weil

By the way, what happened to your hand?

Peter Diamandis

I just had an RFID chip implanted today.

Kevin Weil

As one does.

Peter Diamandis

As one does. I was over at Frontier Towers.

Kevin Weil

That's amazing.

Peter Diamandis

Over at Frontier Towers, a guy named Cass runs their longevity and biohacking. I had this done. If you can feel it right here, there's a little chip in there. It's cool.

It's made by a company called Dangerous Things, which is great marketing, right?

Kevin Weil

Yeah.

Peter Diamandis

But this is the more advanced one. I put it in so I can open my Tesla with it and wave in front of the elevator.

Kevin Weil

Wow.

Peter Diamandis

Check out with Apple Pay.

Kevin Weil

Exactly. I want my—actually, I will not put my crypto keys on this, because then my hand is at risk.

Peter Diamandis

Yes.

Kevin Weil

Yeah, yeah, yeah. The hammer attack. The $5 wrench attack. Exactly. I'm curious. Wait, how big is the chip?

Peter Diamandis

Oh, yeah. You want to see the photos here? It's—

Kevin Weil

Bigger than you would have thought.

Peter Diamandis

So this is what it looks like, right?

Kevin Weil

Okay.

Peter Diamandis

They actually put this giant needle in your hand, about the size of a straw. They make some space, and then they slip it in.

Kevin Weil

Yeah. Nice.

Peter Diamandis

It'll heal in there. I'll put this up on the video.

Kevin Weil

And how long will the batteries last?

Peter Diamandis

Oh, it's not battery-powered. It's near-field, so basically an electric current comes in, right? So anyway, that was fun. It was an hour ago.

So, you guys said to keep conversations internally compressed. I have to believe that you—I mean, OpenAI is the most progressive, but between Gemini, Anthropic, Grok, and everybody, it used to make you wonder whether there would be a hard takeoff and one company that just takes off across everybody else. It looks like there's really a leveling up continuously across the pack. Do you ever look at what the other products are doing and say, “That's a great idea,” or, “Let's not go there”? How do you—

Kevin Weil

For sure. There's so much innovation happening, right? There are a lot of smart people within these walls, and there's a lot more smart people outside of these walls. So we watch what other folks are doing. We also look at how people are using our own models and get a lot of feedback about things they wish they could do and things they love about our models. I kind of think of that all as part of the feedback that you take in.

But what matters at the end of the day is whether we're executing, because I think we have the best people in the world. We have a very clear mission. People are here because we want to build AGI and we see a path to it. So, at the end of the day, it's about our own execution, but I think we'd be crazy if we weren't watching what's happening and trying to learn from it.

Peter Diamandis

So, I met your EA outside and I asked him, “Has Mark Zuckerberg called recently?”

Kevin Weil

Mark's after researchers.

Peter Diamandis

Oh, that's got to be crazy.

I would love to ask you some questions that will help. We have this huge ecosystem of startups that are building AI companies. You were there when the internet was everything. If you're trying to build an internet company, it's pretty obvious that the TCP/IP stack isn't going to change. You're going to be right on top of it.

Compare that to AI today, and you're like, wow, it's such a moving target that you're building on top of. And so, if you can help the teams in any way understand, are we going to build our own Windsurf- or Cursor-type capability, or is that available? Are we going to—

Kevin Weil

Meaning, where is OpenAI not going to go that leaves space for the entrepreneurs to go?

Peter Diamandis

Exactly. Exactly.

Kevin Weil

And I'll give you a case study. Mark Gorenberg, the chairman of MIT, really wanted Sam to come to campus and meet with the president, Sally Kornbluth. And Sam said, “Yeah, I'll come and do that if you introduce me to 50 startups back-to-back.”

So Mark set it all up, and they did the 50 back-to-back meetings. And then Sam was on stage.

Peter Diamandis

That's awesome.

Kevin Weil

But Sam's genuine Y Combinator background means he absolutely cares about the startup economy thriving on top of OpenAI. So the mission is right in alignment. But then knowing what's going to happen next is harder than anything I've ever seen before.

Peter Diamandis

Yeah. So any insight you can give on what's your advice to entrepreneurs wanting to build on top of OpenAI? We've seen a lot of OpenAI wrappers in the past. What's your advice?

Kevin Weil

So here's the context that I think is most important. Every single product we use, every single service that we use today, every single device that we use—all of them, practically all of them, were pre-AI. Any of the ones that are at scale were pre-AI.

Peter Diamandis

Mhm.

Kevin Weil

They're all going to be reinvented.

Peter Diamandis

It's like adding electricity to every mechanical thing.

Kevin Weil

Exactly. It's a really good analogy. Everything is going to be reinvented.

Sometimes the incumbents are going to figure out how to do it themselves. If history is any guide, not many of them are going to figure it out, right? And that's a huge opportunity for startups, for entrepreneurs, because we're in this complete transformation. So that's super exciting. I think it's an awesome time to be an entrepreneur.

Certainly, we're going to try and do some of that reinvention ourselves. No matter how big we get or how many things we do, I mean, everything is different, right? And it's not just tech; it's also the way you interact with your doctor. Material science is different. So we're only going to be able to do this much.

What we say to the developers is, if you're building right at the edge of what the models can do, you're so far at the bleeding edge that the models can't quite do the thing you want, but you can just see little glimmers of hope, that's a great place to be building. In 2 months, we're going to launch a new model, and the thing that you've been just barely getting to work is going to sing. That's going to be awesome.

If instead what you're doing is patching a problem with the current model and you're afraid of the next model launch that we're going to come out with because it's going to—

Peter Diamandis

Disrupt.

Kevin Weil

It's going to obviate the need for the thing you're building. Don't do that. Assume that the models are going to keep getting better at a crazy pace, and build something that's just at the edge so you're excited about the next model, because it's going to make your thing awesome.

Peter Diamandis

Yeah. Aravind Srinivas over at Perplexity calls that AI-complete. When he was designing the Perplexity business plan, it had to be AI-complete, which I think he just made up, but it means exactly what you just said. It rides the wave. It doesn't get swamped, by the way.

Kevin Weil

Yeah, that's a much more concise way of saying what I was saying.

Peter Diamandis

I have a question I've been dying to ask you.

As the chief product officer, what does AGI look like as a product?

Kevin Weil

That's a good question.

Peter Diamandis

Does it just look like ChatGPT, but it can do anything better and faster?

Kevin Weil

Well, chat is such an interesting format because it mimics how humans communicate. How do humans communicate? I can text you, we can write on a keyboard, but I can also talk, and we can look at each other and talk. A chat interface gets at that full generality, so it's really powerful as a backdrop, but it's not the only interface.

I think when you have AGI, your model is going to be creating UI in real time, on the fly, to build the most economical, sensible solution for the thing that you're trying to accomplish. You're going to have all kinds of software being created and thrown away because you can just produce it again in an instant.

Peter Diamandis

Yeah, that's the vision. When you're just talking to it, it's really obvious that it's going to be super empathetic. But then you start to see the video generation, and you're like, "Oh, wait. Now my mind is blown," because it can create scenes for me in real time. It can create images and entertainment, but also, if I'm designing something, it can, like JARVIS, create the design in thin air.

And so now I'm thinking, okay, Jony Ive is going to be working on something.

Kevin Weil

Let's talk about that for one second. I always come back, by the way, to Ender's Game, if you've read it. You know, the jewel in your ear: It can see what you see and hear what you hear, but it's also superintelligent and connects back to the information across the galaxy.

Peter Diamandis

Yeah.

Kevin Weil

That's kind of, I think, where we're going.

Peter Diamandis

Yeah. They lost that in the movie, actually. You have to read the book.

Kevin Weil

Yeah. So I think we're going to this multimodal world where there's video capturing everything and you're always being seen in your environment, and where AGI becomes anticipatory. It knows what you likely want, right? It's making the world automagical. That's the term I use for it.

Peter Diamandis

I'm still trying to dig into what AGI as a product from OpenAI looks like. Is it fully anticipatory? Is it just doing things you don't expect it to be doing? I guess you could turn that feature on.

Kevin Weil

Yeah. I think it's also there, and I don't even know that we need to be fully at AGI for this. I think you want the product to be taking care of everything, whatever AGI is.

Peter Diamandis

Yeah, and just doing—I mean, you think about the number of things that you do every day. You wake up in the morning and you have a whole bunch of emails. A lot of those are mechanical. Some of them are ones where you really want to type out every word of the response, but most of them are mechanical. They're scheduling and responding. I want my AI to have already done all of those for me before I wake up.

Kevin Weil

Yeah.

Peter Diamandis

Maybe every once in a while I need to say, "Yeah, that's right," or, "I want to edit that sentence." But mostly it's just like, "Be done."

Kevin Weil

Yeah.

Peter Diamandis

You're going to have that across everything. How much of your browsing is semi-mechanical? You have to get the thing done.

Kevin Weil

Mhm.

Peter Diamandis

But you actually don't care about reading the words on the page, waiting for things to load, and clicking links. You're just trying to get a job done. All of that should be gone.

Kevin Weil

Yeah.

Peter Diamandis

But what do you think about the future of media? One thing that's really interesting about media in general is that everything competes with everything else. You don't really think of sports as competing with news, but it does because it's all competing for mindshare. Clearly, AI is going to take a huge amount of our mindshare because it's so engaging, and that has to come from somewhere.

Do you guys think through how this is going to affect internet media, television, movies—all of that?

Kevin Weil

Attention economy? Sure. It probably becomes hyperpersonalized. There are good and bad things about that. There's something magical about every single person in the United States watching the same 2 news shows at night, and we don't have that anymore.

Peter Diamandis

Yeah.

Kevin Weil

There's something cool about the shared experience: The same movie is in the theater for all of us at the same time, and we can talk about it. But there's also something pretty cool about a movie being totally personalized to me and being the only one that matters to me. You have to think that kind of thing starts to become possible.

You also probably go to extremes. The other extreme is that you value the thing created by humans that much more. Sports are the last business model in television that still really works well.

Peter Diamandis

Well, you're an ultramarathoner. You'll still be doing that. I think what's not going to change in the next 10 years—I mean, we can list so many things that will.

Kevin Weil

Yeah. Well, I think humans are still going to be humans, right? Us connecting in person is really going to matter. The feel of this is very different—us doing it here, sitting around a table, relative to us doing it through Zoom or something. That's still going to matter.

A handwritten note is still going to matter. Personal connection is still going to matter.

Peter Diamandis

I would think it's going to matter a lot more, actually. You mentioned earlier that you just love having everybody here in the building. It's just different.

Kevin Weil

Yeah. And since you have so many AI force multipliers, you actually don't need regional offices all over the world. So the relative importance of communication inside the team is probably higher than at any company you've been in before.

Peter Diamandis

I could run my Abundance Summit virtually throughout the year, but we're together for 5 days in March.

Kevin Weil

Yeah.

Peter Diamandis

And those 5 days are just magical in terms of the energy.

So what do you guys think? What stays the same, and what does AGI look like? One thing that occurred to me after the GPT-5 launch is that, up until this point, my kids were working with GPT-2 to write content, then GPT-3 and GPT-4, and they've been watching this crazy progression.

Now it's pretty clear that the AI community is going to try not to change too many things too quickly. Let the AI become superintelligent. Let it solve healthcare, but don't let it go rampaging across San Francisco, tearing down buildings, because that's actually creating more problems than good.

It's becoming clear to me that the AI community is really small as a fraction of the world, and the last thing it needs to do is change everything tomorrow, even though it potentially could. So it's starting to feel to me like the physical world won't change as much as people might think, while the virtual world will change at warp speed.

Kevin Weil

Mhm. I think the importance of purpose is still going to be right, because one of my biggest concerns about the endpoint here is when everything is being done for you or can be done for you, and you can just sit back and be a couch potato.

The thing that's going to make you live in a Star Trek universe, not a Mad Max universe, is purpose. That importance—

Peter Diamandis

Yeah.

Kevin Weil

Yeah, I completely agree with that. I always have a hard time with the futures that people paint when it's like, "Oh, we're all just going to sit back, eat grapes, write poetry, and receive our UBI," because I don't think that's what drives us as humans.

If you go back and look, we don't have to cross the United States in a covered wagon. We don't have to hang our laundry on clotheslines as much anymore. If you go back 100 years, people would look at our lifestyle today and say, "Oh, my God, everything is done for you. What are you talking about?"

But of course, we take those time savings and go do more interesting stuff with them.

Peter Diamandis

Yeah, it's always going to be true.

Kevin Weil

It's all about time savings, right? Every single human—8 billion of us—has 7 days in a week, 24 hours in a day, and 365 days in a year. What you can do with that time defines wealth, success, impact, everything.

Peter Diamandis

Yeah.

Kevin Weil

And AI is the greatest time multiplier, period. I think it's one of the most powerful parts of human nature that we all strive for something bigger than ourselves. We want to leave the world a better place than we found it. That's so innate to humans. AI is not going to change that; it's just going to supercharge it.

Peter Diamandis

So, without asking you now to tell me what Jony Ive is building for you, we are going to interact with AI in different ways. How do you think about that future, besides the fact that it's been typing and now voice is amazing?

Kevin Weil

Mhm. Yeah. I mean, the jewel in your ear is the model in so many ways, because you don't want to take ChatGPT—it's just one of many AI products that we'll all use.

You don't want ChatGPT to just sit inside a phone app or a browser, waiting for you to go there. You want it to be wherever you are. You want it to effectively see what you see and have all the context that you want to give it, so that it can be smarter and act more proactively for you. It shouldn't wait for you to open a phone app to do stuff for you. It should be constantly chugging away behind the scenes, taking action for you and saving you time.

Peter Diamandis

That's going to be much more the future than today, where you go to a website and ask it questions. Can we come back to GPUs for just one second?

Kevin Weil

Yeah.

Peter Diamandis

And let's talk about Stargate, too.

Kevin Weil

Yeah. Stargate—Chase Lochmiller is walking around like he's our MIT classmate, an alum. He's walking around Abilene, Texas, with Sam, and there's concrete and pipes and stuff everywhere.

Peter Diamandis

Yeah. It's not up and running yet, right? None of the functionality is delivering yet. Getting closer? Is it getting closer? Is it very soon, or is that all secret?

Kevin Weil

Yeah. I don't remember what's public and what's not, but we're very excited for that to come online because we could use all the GPUs we can get, I'm sure. All we know is that, in the Oval Office, there's a $500 billion budget. It gives us a sense of the budget.

Peter Diamandis

Why stop there?

Kevin Weil

Yeah. Why stop there? I don't think we will.

Peter Diamandis

No.

Kevin Weil

Yeah.

Peter Diamandis

I just got back from Brazil. I was there delivering 5 keynotes and talking to entrepreneurs and builders. They're all asking, “What will sovereign AI look like for us? How do we make sure we're not passed over?” The same question is probably being asked in India and other parts of the world. Speaking to those entrepreneurs, they're extraordinarily passionate. What's the advice to the leadership in those regions, in South and Central America?

Kevin Weil

Part of this is why we try to offer everything. ChatGPT is a free product.

Peter Diamandis

Yep.

Kevin Weil

Download it from the App Store or on the web. You don't even need to sign in. You can just go and get access to the greatest intelligence the world has ever seen for free on your phone at all times.

Peter Diamandis

Yeah.

Kevin Weil

And, as you said, we launched GPT-5. We cut prices dramatically. If you look at prices today relative to what GPT-4 cost when it came out—

Peter Diamandis

Even as models have gotten way more intelligent, they've gotten 100 times cheaper.

Kevin Weil

You're on this crazy cost curve, and we want to keep going because the more intelligence we can put in people's hands—

Peter Diamandis

The more they can do with it. Do you feel like that ever would have happened if OpenAI hadn't forced the timeline? You were at Meta, you've been at Twitter, and you've seen it. If this stuff had stayed inside Google's labs and OpenAI hadn't come into the world and said, “Hey, we're going to make this available right now,” because now Google is throwing it out there the same way OpenAI has—but that's because it's a reaction.

Kevin Weil

It's a reaction, right?

Peter Diamandis

So where do you think we would be if OpenAI hadn't existed? Would it all be hidden somewhere still? We wouldn't even have access to it?

Kevin Weil

We wouldn't be where we are, right? Competition is great. As a friend of mine likes to say, capitalism is undefeated. It's amazing to have competition. By the way, Google is competition for us. They're moving very quickly and building good models. So is Anthropic. So are a bunch of other players. It's motivation for us to also go faster. Consumers and businesses benefit, which is great.

Peter Diamandis

Do you remember the moment in your life when you first said, “I want to go work at OpenAI. I need to go get in that building, meet Sam, and get there”?

Kevin Weil

Yeah. I've known Sam for a bunch of years. We're not super close, but we were friends. We could text, that kind of thing. I would call him whenever I was thinking about going from one job to doing something next, because he's such an incredible thinker and futurist, and he's connected.

He's highly connected, but he's also involved in interesting things. I'm a physicist by background, and before my previous job, about 4 or 5 years ago, I called him. He introduced me to some companies building fusion, and I almost went and worked at one of those companies.

Peter Diamandis

So you could literally be working on magnetic containment for a fusion reaction somewhere completely unrelated.

Kevin Weil

Pretty amazing, right? Like, 37 venture-backed fusion companies. I would have never imagined it.

Peter Diamandis

Are there 37? Yeah. I love hearing that. By the way, I'm so glad there has been a renaissance in deep tech and investment.

Kevin Weil

Yeah. There are venture-backed rocket companies and fusion companies. It's like, okay, this stuff is becoming real.

Peter Diamandis

Yeah, we need more of it.

Kevin Weil

Exactly.

Peter Diamandis

How can you possibly keep up with the speed of change? I think, in your role, trying to keep up with how rapidly the technology is progressing, how do you think about that?

Kevin Weil

Part of it is staying as connected to our research teams as we possibly can.

Peter Diamandis

How big is the research team here?

Kevin Weil

I don't know.

Peter Diamandis

Order of magnitude?

Kevin Weil

A few hundred. Yeah.

Peter Diamandis

Something like that. It's very independent—it seems very independent, right? People can work on a project and begin to launch features.

Kevin Weil

Yeah. There's kind of a spectrum. Parts of the research team are super academic. They're doing novel research that may or may not work out, may or may not see the light of day for a while.

Because this is such a new field, there's a lot to discover. You're not just productionizing things that somebody else figured out. You have to figure this stuff out from scratch. Then there are parts of the research team that are much closer to product, where you're taking a model and helping to post-train it and making sure that it's good at certain things that are really important for our customers in various shapes.

You kind of have everything in between. One of the things that I think differentiates OpenAI is how closely product, engineering, and research all work together. There are parts of research that are very separate, but when we work closely together, we get a tight iteration loop: building model capabilities, turning that into a product, getting feedback, taking that feedback back, improving those model capabilities, and iterating that way. That's how we built deep research. That's how we build agent functionality. The best stuff comes from that.

Peter Diamandis

In your role, here's research—ultra-science—then product features and functions: this voice, this button. Over here, you've got data centers. Do you span all of that as CPO?

Kevin Weil

No, just the products that you use. Think about the API.

Peter Diamandis

So Mark—Mark is overseeing all the—

Kevin Weil

Yeah. Mark leads research. Greg leads everything—data-center infrastructure, scaling, all that kind of stuff.

Peter Diamandis

Got it.

Kevin Weil

Yeah, there’s a lot going on. We divide and conquer.

Peter Diamandis

The feature I’m looking forward to is my BCI connection.

Kevin Weil

Oh, yeah.

Peter Diamandis

A friend of mine is involved in Merge, which was just announced—or leaked, I guess. We have 100 billion neurons and 100 trillion synaptic connections. Being able to have a feature set that lays on top of my neocortex—that’s exciting. Do you actually think that far ahead in your conversations?

Kevin Weil

Is Merge doing something invasive, by the way, or is it sitting on your scalp? How are they doing it?

Peter Diamandis

I know the details of the company, but I don’t know what I can say about it.

Kevin Weil

Okay.

Peter Diamandis

There are a multitude of companies that are invasive, and a number that are external. There are also those that are subcranial but above the dura of the brain.

Kevin Weil

Lots of different layers there. I’m super excited about that, too.

Peter Diamandis

Oh, yeah. I would do it in a heartbeat. Would you?

Kevin Weil

No. Absolutely not. But that’s only because I use AI literally 6, 7, 8 hours a day, and it’s getting so far ahead. The original vision was, “I need to increase the bandwidth of communication between my brain and this really slow laptop.”

Peter Diamandis

Yeah.

Kevin Weil

But now the AI moves so fast that it’s clear I can give it a concept and it can just run with it. I haven’t had the next concept yet. Increasing the bandwidth between my brain and the laptop no longer matters. The AI is just way too fast.

Peter Diamandis

Alex Wissner-Gross, who’s one of our friends and partners, was on my stage at the Abundance Summit. He was talking about the importance of coupling: If AI is taking off in a real way and humans are linear, can we couple with that? That coupling is going to need to have this kind of interface. So here’s the question to you: Would you do a BCI coupled to OpenAI?

Kevin Weil

I would personally do it. I would personally do BCI.

Peter Diamandis

Yeah. When it’s safe, I’ll do it in a heartbeat.

Kevin Weil

Yeah. I tend to do experimental things in the first place.

Peter Diamandis

I was going to say, says the man with an RFID chip in his hand.

Kevin Weil

No, I would totally do it. I think it’ll be amazing when you can access the world’s information not by typing into one of these things, but at the speed that you can think. Then you have this—I mean, we will be a different species. It’ll be super cool.

Peter Diamandis

Yeah. We are going to speciate. There’ll be those who decide, “I like it the way it is. I don’t want that in my head,” and others who say, “I want to see this. I want to understand quantum physics. I want to have infinite knowledge. I want to be able to interface with it.”

Kevin Weil

Yeah, I understand what you’re saying completely. You’ve got the Amish over here and the technophiles over there.

Peter Diamandis

That’s not what I’m saying. I’m saying that when you surround me with monitors—

Kevin Weil

Yeah, and it’s creating things on those monitors at the speed of AI right now.

Peter Diamandis

There’s no bandwidth increase beyond that. I can’t absorb any more information than that anyway. I’m at the limit of what I can think about.

Kevin Weil

But maybe not with your eyes, and maybe not when you have to type it into a thing and then wait for it all. It’s really hard to beat.

Peter Diamandis

We’ve talked about the fact that visually, you can look at an image and tell whether the data is correct or not, right? If you have a text, you have to read every line. Anyway, we’ll see. We’re going to find out.

Kevin Weil

Well, I guess you and I will be the A/B test, or we’ll be the A and he’ll be the B.

Peter Diamandis

You’ll be the A. If you’re still alive, you’ll be the B.

Kevin Weil

If I’m still alive, yeah.

Peter Diamandis

Yeah. Peter’s having it done tomorrow. I might wait a year, and then we’ll see how that goes.

Kevin Weil

As long as you can upgrade. You don’t want to get version 1 and then—

Peter Diamandis

It is true.

You’ve worked at some amazing companies. I love what Planet is doing as well, putting a layer of imagery on top of the planet. I can’t imagine that the speed of iteration of products is anyplace faster than it is here.

Kevin Weil

Nothing in my experience compares to OpenAI.

Peter Diamandis

Yeah.

Kevin Weil

I thought Facebook and Instagram, when I was there, moved quickly—and they do. But nothing compares to OpenAI.

Peter Diamandis

You’re doing a $10–20 billion capital raise at a $500 billion valuation, something like that? I’m just reading the news. I don’t want to—

Kevin Weil

No, you’re just reading the news. You tell me.

Peter Diamandis

Okay. Do you feel like there’s a vulnerability at the raw compute and budget level? Google has about $100 billion of free cash flow. Amazon is huge, and they have massive data centers in advance.

Kevin Weil

That’s why we’re so focused on building out our own capacity. They’ve got big data centers. They’ve been doing this for a long time, and they’ve built great technology. They also support a whole host of things. Google builds a lot of products that we all use.

Peter Diamandis

Mhm.

Kevin Weil

We build one product. We have one mission. Building AGI is existential for us. I think that’s an advantage.

Peter Diamandis

I love that simplicity. It really goes back to Google’s earliest roots, where they were just a search engine, just a box on a white page.

Kevin Weil

Mhm.

Peter Diamandis

Right.

Kevin Weil

There are lots of places you can go work as a researcher that will pay you a lot of money. But if you want to work at a company that is maniacally focused on getting to AGI, you work at OpenAI.

Peter Diamandis

Yeah. I asked GPT how many core researchers there are at Google, just in the AI team. Jeff Dean, Demis Hassabis—it’s like 5,000 or 6,000, roughly, assuming GPT is right. I don’t know. It’s a much bigger force, but it feels like they’re working on protein folding and a lot of other things. They’re very dispersed in what they’re working on, and it wasn’t clear to me how many were just working right down the fairway toward AGI relative to OpenAI.

With the GPT-5 launch, it feels like the coding completely caught up with—or passed—everything else out there at a fraction of the price. The momentum seems to be great, but the headcount has got to be smaller.

Kevin Weil

Yeah, but that’s every startup’s advantage: agility.

Peter Diamandis

It’s all about agility.

Kevin Weil

Yeah.

Peter Diamandis

We’ve had some conversations on the Moonshots podcast about benchmarks. How much do you think about benchmarks? It looks like many of them are getting saturated, and we’re reinventing benchmarks. You’ve created a healthcare benchmark that you use, and one of the questions we talked about was creating benchmarks around AI’s ability to solve grand challenges and fundamental problems.

Kevin Weil

Yeah, I love that idea.

Peter Diamandis

We called it a series of abundance benchmarks. You measure all of the AIs against something like that. It’s a proof of work instead of a proof of stake.

Kevin Weil

Yeah. It’s a really interesting thing, and it’s really important, too, because we’ve had all of these different benchmarks that we’ve used for the last bunch of years in AI, and they’re almost all saturated now.

Peter Diamandis

That’s crazy.

Kevin Weil

It shows how good the models are getting very quickly, but it also means that we need harder benchmarks. There are a bunch of these things that are FrontierMath and ARC-AGI, but increasingly we’re also looking at what I’ll call softer benchmarks, which is a challenge in itself. We have HealthBench, and we’ve got benchmarks around whether you can build a financial model for a company.

When you talk about AGI, one of the ways that we define AGI is: Can you do economically valuable tasks?

Peter Diamandis

Sure.

Kevin Weil

You look at what people are doing every day if they’re a doctor, a lawyer, a banker, and so on, and you ask whether you can do those tasks.

Peter Diamandis

Yeah.

Kevin Weil

But they’re also softer. It’s not obvious. There’s not one way to build a model for a company or a financial model. You could do it in a bunch of different ways. If you’re doing a math problem, there’s one right way, and you can grade it very easily.

At the same time as models are getting smarter, we’re saturating the things that are easy to grade. We’re moving into these harder problems that are also a little bit tougher to grade. They don’t have to be deep science or anything. You can also look at creative writing. How do you grade creative writing? There’s way more than one way to write a story about any particular thing.

Peter Diamandis

So true. And increasingly important, right?

Kevin Weil

Yeah. The stuff is running wild in areas where the evals can be built, and it can’t run wild in areas where there’s no eval because it’s just unable to self-improve. But anyplace you can score yourself—

Peter Diamandis

I saw a recent metric on its ability to predict the future.

Kevin Weil

Okay.

Peter Diamandis

I found that fascinating. I mean, this goes back to Asimov’s work in the Foundation trilogy.

David Blundin

Can you—are there enough clues out there that we can gather to predict what society is going to do in the next 12 hours, 12 days, or 12 months?

Speaker 2

Uh-huh.

Speaker 1

Yeah. Peter wrote a book, The Future Is Faster Than You Think. I don't know if you read it. It's my favorite of Peter's books. There are many great Peter books, but it's my favorite because it talks specifically about convergent technologies that nobody's figured out how to glue together.

Speaker 2

Uh-huh.

Speaker 1

Normally, there'll be 2 or 3 things: We have lithium batteries, we have control systems—let's make EVs. Two or 3 things.

Speaker 2

But with AI, there are so many things that got cracked in the last 6 months.

Peter Diamandis

Yeah.

Speaker 2

They haven't been glued together yet into products, services, or whatever. It's just accelerating. There's this flaw in human psychology where something is mind-blowing, and then you take it for granted within 3 days.

Peter Diamandis

Yeah. Yeah. Totally.

Speaker 2

And then you miss the center. You miss the mental breakthrough that says, “Wait, I can glue that together with this.”

Snapchat is my favorite example. That company had a $20 billion valuation. John Jarve from Menlo Ventures has a favorite story about how Evan Spiegel was a roommate of his son—or a fraternity brother of his son—at Stanford.

Peter Diamandis

Yeah.

Speaker 2

He got a first look at it and said, “That's just too simple. There's no way.”

Peter Diamandis

Yeah.

Speaker 2

Then it was a $20 billion IPO. It was probably his biggest investing mistake, maybe the only big mistake of his investing career. But it was like, look, the phone has a camera, and the camera is high-resolution. This functionality is just the assimilation of texting and a camera.

Peter Diamandis

Yeah.

Speaker 2

That's all you had to realize. But those 2 things kind of grew in parallel, and people didn't stick them together. Now you've got 40 things you can do with AI that you couldn't do just a few months ago, and they can be glued together in all these different ways. All the combinatorics are interesting products waiting to be invented.

Speaker 1

Yeah. And I think where AI is today, if it just stopped—if progress stopped at GPT-5—you'd still have a complete transformation of society over the next decade.

Speaker 2

So true.

Speaker 1

But it's not going to stop. It's going to keep going, and it's probably going to keep going on the same essential exponential.

Speaker 2

Yeah.

Peter Diamandis

Yeah. I should update the book. Well, the next book that's coming out in March, which I'm releasing at the Abundance360 Summit, is called We Are As Gods.

Speaker 1

And it's a realization that what we do between breakfast and lunch is godlike to our ancestors.

Peter Diamandis

Right. I mean, we take it so much for granted, but our ability to manifest the future, to literally know anything anywhere, and to connect with anybody for free—I mean—

Speaker 1

Yeah.

Peter Diamandis

It's crazy. AI is one of the first things. The transformation that it'll bring about—you've got to think it's just the beginning. You grow up and think back to when your grandparents grew up, and you're like, “I'm so glad that I'm living at the time that I'm living.”

Speaker 1

Best time to be alive.

Peter Diamandis

AI is the first time you look at what's happening in biotechnology, brain-computer implants, and all these other things, and it's the first time I've been like, “Man, I am jealous of my kids.”

Speaker 1

Yeah.

Peter Diamandis

Because the year 2060 is going to be so cool.

Speaker 1

Well, you'll be here.

Peter Diamandis

I mean, I hope I'm here. But, I mean, come on. This is why—the work you're doing, I hope I'm here in 2100. It's the reason for me. I grew up passionate about space. I wanted to see space play out. NASA never did it. Elon has gone heads-down, and Bezos and others are working on it, so it's fantastic.

But when I read Ray Kurzweil's book, The Singularity Is Near, it was like, “Holy—this convergence of these technologies is going to blow everything away and accelerate it.” I want to see the interstellar missions. I want to see what uploading consciousness looks like. I want to see all of these things, and I don't think there are any limitations beyond the laws of physics as we know them on what we're going to be able to do.

Speaker 1

So—

Peter Diamandis

And it's going to keep accelerating. Yeah.

Speaker 1

Talking about your twins, how old are they now?

Speaker 2

Eight, almost, and mine are 14.

Speaker 1

Okay.

Speaker 2

And so I think about their future. Speaking to the parents listening, how do you think about educating kids today in an age of AI? I'm just livid about the schools that are saying you can't use AI at all in school.

Speaker 1

Oh, yeah.

Speaker 2

Listen, my kids are going to be using it the moment they graduate, everywhere.

Speaker 1

And I think one of my challenges is that if you give an eighth grader eighth-grade homework to do with AI, it's meaningless. But if you give them graduate-level work—go design a starship and figure it out—I mean, that's amazing. That's what we should be doing.

Speaker 2

Yeah. Assume that ChatGPT exists and use it to make students, A, learn how to use the technology, but B, go deeper than they ever would have been able to in a classroom setting.

Speaker 1

Yeah.

Speaker 2

They should emerge stronger from using AI with their education.

Speaker 1

Yeah.

Speaker 2

Actually, the best book that I've read on this, at least, is by Ethan Mollick. Have you ever read his book? It's called Co-Intelligence, I think.

Speaker 1

Okay, great.

Speaker 2

It's a short book. It's about a 90-page read—you do it in an hour. He's a professor at UPenn; I think he's at Wharton. It's just a great, quick read, super practical.

Speaker 1

For you guys, it's just a great, quick read—super practical.

Speaker 2

He's incorporated AI into his class and has changed the way he teaches to assume that everyone is using AI. He uses that to take them much deeper into a topic than they would have been able to go and raises the expectations on the final work product as a result.

Peter Diamandis

Exactly right. I think that's true for all of society. As we get to—let's forget about AGI and talk about ASI, where intelligence is a millionfold or a billionfold more capable—what do we do with that? Humanity needs to go for these grand challenges. We need to have these epic visions.

Speaker 1

I couldn't agree more.

Peter Diamandis

Otherwise, we will fade into insignificance.

Speaker 1

Yeah. There's another book that I really loved by Robert Zubrin, The Case for Space.

Peter Diamandis

Yeah, I know. Bob has been a friend for many, many years. He makes this point very eloquently in the book about the human need for frontiers, which is one of the reasons space is so important. The bottom of the ocean is still kind of a frontier on Earth, but there aren't that many frontiers left.

Speaker 1

And when there's a frontier, no matter how you screwed it up wherever you live, on the frontier you can start again.

Peter Diamandis

Yeah. Yeah. Space is about innovation, refreshing, and renewal.

Speaker 1

So when do you think we will be multiplanetary?

Peter Diamandis

We'll see. Hopefully, there's a Starship 10 launch this Sunday. Starship will get us to the Moon on a recurring basis by, hopefully, the end of 2026—definitely 2027. Then Mars—my guess is boots on Mars by 2030, but they'll be Optimus robot boots on Mars.

Speaker 1

And you're limited in your iteration speed there because you only have a—

Peter Diamandis

You are, but you launch only once every 18 months.

Speaker 1

But you're going to send robots in advance. You'll send the robots to the surface of the planet. You'll send robots to the asteroids to mine materials.

The way I think about it is that everything, at the end of the day, is one way or another driven by economics. Everything we hold of value on Earth—metals, minerals, energy, real estate—is available in infinite quantities in space.

Peter Diamandis

Yeah.

Speaker 1

The Earth eventually will be a crumb in a supermarket filled with resources.

Peter Diamandis

Right. And that's a beautiful thing.

Speaker 1

How's that going to evolve? Mars is an obvious target now. Everyone's been working on it.

Peter Diamandis

I like the Moon better personally.

Speaker 1

Better than Mars?

Peter Diamandis

Yeah. I mean, as a stepping stone or as a place to set up, go start a city. There are these lava caves under the surface of the Moon where you get protection from radiation. If you fill them with air at 1/6 gravity, you can strap on wings and fly.

David Blundin

That's awesome.

Peter Diamandis

I do this on my phone. It's so much fun.

David Blundin

How big are the caves?

Kevin Weil

Is there a massive cave system?

Peter Diamandis

There are huge caves, yes. You could build—you could have thousands, maybe millions, of people in there.

David Blundin

I don't know about millions. I can easily imagine thousands or tens of thousands. And, of course, there was the recent announcement that NASA is planning to put a nuclear reactor on the South Pole of the Moon.

Kevin Weil

Mm-hmm.

Peter Diamandis

So cool, right? In order to literally electrolyze the ice that's on the South Pole and create water and oxygen.

Kevin Weil

I mean, we're finally living into that science fiction future, and it's accelerating.

Peter Diamandis

Yeah. And hopefully fusion soon.

Kevin Weil

And hopefully fusion will solve a lot of problems here on Earth and open up new vistas for space.

Peter Diamandis

Sam is, like, what, 2028 for Helion, I think, is the current target date, and CFS is coming online in 2030, thereabouts.

Kevin Weil

It's amazing how a lot of these things bust loose when there's a single person who's just got an idea and is incredibly passionate about it. Now, with AI as a force multiplier, the odds of those passionate people becoming empowered and actually achieving that mission are so much higher.

Peter Diamandis

Mhm.

Kevin Weil

Than they were 2 years ago.

Peter Diamandis

Yeah. And then you get the right density of startups, and you have competition and different approaches, the ability to learn from different approaches, and the probability of success grows.

Kevin Weil

And people are just trying to outdo each other and have fun. You could be in this future where it really is a Star Trek future instead of a Mad Max future. It's like, what ambitious, super-cool thing can you go for?

Peter Diamandis

Yeah. It used to be just a small number of people. It used to be just the kings and queens and robber barons who could do anything impacting people. Now it's anybody using these tools.

Kevin Weil

Yeah.

Peter Diamandis

It's amazing. So I have a question, as a chief product officer: Who names your models?

Kevin Weil

Ouch.

Peter Diamandis

Because it's like, okay, I'm trying to understand the logic and the rhyme and reason.

Kevin Weil

Yeah, yeah, yeah. We take a lot of well-deserved flak for the naming of our models.

Peter Diamandis

Sorry to give you a hard time, by the way.

Kevin Weil

No, it's totally fair. We lean into it. The fact that we had both o4-mini and 4o mini at the same time is something to be proud of.

Peter Diamandis

And then make the user choose, if they have any. On the other side, there's just ChatGPT simplicity.

Kevin Weil

Yeah, it's good.

Peter Diamandis

Yeah.

Kevin Weil

You know, in all seriousness, we deserve all sorts of flak for that, so it's totally acknowledged. But one of the reasons that you have that is the philosophy of iterative deployment. We could have waited. You have some new capability, like reasoning, for example. Up until we launched GPT-4o, it could do a bunch of new things, like you could speak to it rather than just type to it.

Peter Diamandis

It's amazing.

Kevin Weil

Yeah. We'd been working on the ability for models to reason—not to answer right away, but to actually think through the problem, try a bunch of different approaches, test them out, the way you would if I asked you to do a crossword puzzle or a Sudoku or something. You don't just spout out the answer. You've got to try things. You're like, “Okay, if that was a 6-letter word starting with A there, but it's got to have a B and this. Okay, it could be that. No, not the A.” You're testing hypotheses, refuting some of them, accepting some of them, and then moving on.

It took a while to teach the models that. It was one of the big breakthroughs coming from OpenAI. The first model that could do that was o1, and it was easier and faster for us to launch a model that just did that. It wasn't an ideal model to ask for relationship advice or to ask who the third Holy Roman Emperor was, or whatever—all the normal things that you can ask ChatGPT. It was easier for it to be a very specialized model. It was very good at thinking hard about scientific problems in particular.

Launching it as a separate model allowed us to iterate faster and put this thing in people's hands, see what they did with it and what they wanted to do, what its shortcomings were, and where it excelled. So then we iterated a bunch that way. You have GPT-4, o1, o2, and all that, and then you have the mini models, which are smaller, faster versions of those things.

Peter Diamandis

Yeah.

Kevin Weil

And GPT-5 for us was the moment where we brought all of this stuff together and had one experience.

Peter Diamandis

Elegant.

Kevin Weil

But I expect in the future that we'll have other things where we have some new capability and want to test it out faster than we can integrate it into one massive thing. I wouldn't be surprised if we have these kinds of offshoots again in the future, and then integrate them as they work and as we gain mastery over them.

Peter Diamandis

Yeah. Eric Schmidt talks about learning loops.

Kevin Weil

Right. The speed of innovation is the speed at which you get learning: Put it out there, see how people think about it, and get their feedback.

Peter Diamandis

Yeah.

Kevin Weil

And especially with these models, we're all kind of uncovering what they can do. It's especially valuable for us to get them out into people's hands and let people build.

Peter Diamandis

Yeah. The reasoning was just shocking, actually. I don't know if it was true here, too. Nobody prior to o1 would have thought that chain-of-thought reasoning was going to be such a huge unlock.

Kevin Weil

Yeah. I mean, it starts with iterative reprompting, which just worked far better than anyone ever would have guessed. That evolved into o1 and then o3. I guess o2 was oxygen. I don't know what happened to o2—some trademark issue or something.

Peter Diamandis

Trademark.

Kevin Weil

Yeah. I'm sure there'll be other offshoots, because the capabilities are always surprising. Even the best researchers in the world don't know exactly what's going to come next.

Peter Diamandis

So I'm curious. It's interesting, right? We just cruised through the passing of the Turing test and didn't notice.

Kevin Weil

Isn't it amazing? It's crazy. It was held up for 50 or 60 years as this pinnacle, and then we just whooshed by it.

Peter Diamandis

What is the effect? People take it for granted so quickly, and they miss the implications because they're already on to the next thing. “Oh, I took it for granted.” No, you've got to think about what took place. It's right here in our hands.

Will we pass through AGI the same way? It will be like, “Oh, yeah, it happened.” I really think so, because—

Kevin Weil

Arguably, AGI is here. It's just unevenly distributed, right? There are lots of ways where the model is already way smarter than me, and I would never choose myself over the model.

Peter Diamandis

Mhm.

Kevin Weil

And then there are lots of places where the model is definitely not smarter than me yet, and I would choose myself. But over time, I'm kind of staying the same. I'm improving a little bit, hopefully, but the model's getting better a lot faster. The level of water is rising.

I just saw that GPT-5 Pro was measured at an IQ of about 148.

Peter Diamandis

Okay.

Kevin Weil

On the Mensa Norway scale.

Peter Diamandis

Wow.

Kevin Weil

That's impressive.

Peter Diamandis

Yeah. I mean, it won a gold medal at the IMO, right?

Kevin Weil

Congratulations. It was second in one of the programming competitions.

Peter Diamandis

I have to say, it was interesting, right? I looked at the IMO statistics, and the U.S. came in second. I think it's out of 36 points.

Kevin Weil

Yeah, 36 out of 42. The U.S. had between 32 and 36. The Chinese team that came in first had a perfect score across all 6 Olympiad problems.

Peter Diamandis

Wow. Our podcast mate, Alex Wissner-Gross, who's a total genius, has been tracking that IMO score as the benchmark for self-improvement because solving those problems is very highly correlated with iterating through the algorithm.

Kevin Weil

Yeah.

Peter Diamandis

So I think there's a lot of belief now that we're going to go into this really rapid acceleration of capability because of self-improvement, until the GPUs run out.

Kevin Weil

Yeah. Some people think so.

Peter Diamandis

But the electricity runs out first.

Kevin Weil

Well, I think it's interesting because the GPUs are constrained. The electricity won't run out quite yet because we can't make the GPUs fast enough to keep up. But the software improvement is unconstrained, right? There's this period of time, I think it's coming right now, where the algorithmic improvement—

Peter Diamandis

Algorithmic improvement.

Kevin Weil

The algorithmic improvement that we estimated on the podcast at between 100 and 1,000×—recently, there's some new news coming out that indicates maybe it's even higher than that.

Peter Diamandis

I don't know if you guys have a theory on that internally or not.

Kevin Weil

Well, there are now 2 dimensions to how we're scaling intelligence, right? One is through pretraining, which is kind of the traditional thing, and now there's test time. How long are you willing to give the model time to think, and how long can you keep it on track as it does that?

Peter Diamandis

Yes.

Kevin Weil

With o1, o3, and ChatGPT, you get the model thinking for about 60 seconds, and then it'll come back and give you an answer. Most people don't want to wait too much longer than that. Deep Research, though, will sometimes take 20 or 30 minutes to do a bunch of research and iteratively compile everything that it gets, figure out what it's missing, and go back and get more. But there's no reason that you can't have a model think for 2 days, 2 weeks, 2 months, or 2 years.

Peter Diamandis

Andrew Wiles didn't solve Fermat's Last Theorem by thinking for 5 minutes.

He worked on it for 7 years.

Kevin Weil

And we see continued evidence that the longer the models think, the smarter they get and the harder the problems they can solve.

Peter Diamandis

So, I mean, there's this other dimension of scaling that today, at least, we don't see any limits. We see a lot of continued growth there. There's a lot of work going on for custom chip design that's designed by AI for specific applications. Do you have a whole strategy around that? Are you working with, you know—

Kevin Weil

Yeah, that's one of those areas where you can let the model think, and it's a pretty well-constrained problem.

Peter Diamandis

Right. There's a great simulation that you can optimize on, and you can run tests and understand if your layout is better than the previous layout.

Kevin Weil

Yeah.

Peter Diamandis

And the more time you give the models to think about it, the more breakthroughs they make. So I actually agree with you. I think that's one of the areas where we're going to see—I mean, we already are seeing—real innovation. I believe Google has been public about the fact that they've designed their TPUs and improved them with AI.

Kevin Weil

You know, we're doing similar things.

Peter Diamandis

I think if you talk to Noam Shazeer or anybody over there, historically, the TPU team has been way the hell over here and the algorithms team has been way the hell over there. Trying to translate your algorithmic idea into something that the hardware guys understand is just a nightmare. So that's been a very slow process, but now the AI will do it for you. It'll do the chip design right off your software design. That's going to be the big unlock. I don't know if it's this month. It's not—

Kevin Weil

Oh, I think it's already happening.

Peter Diamandis

Yeah.

Kevin Weil

It's already happening.

Peter Diamandis

Do you have a team here doing specifically that?

Kevin Weil

Yeah, we're working on our own chips, and we would be crazy if we weren't also using AI to improve our chip design and layout and everything.

Peter Diamandis

Do you have all the manufacturing and fab figured out, too? Is that—

Kevin Weil

We're working with partners on that, for sure. But, yeah, there's this class of problem where you have a fairly well—it's a specific problem, and you have a way that you can grade it. In this case, it's the speed of the chip that you design.

Peter Diamandis

And in problems like that, where you have these well-specified problems, you can just iterate and iterate and iterate and apply arbitrary amounts of compute. So far, I think we're going to see arbitrary amounts of improvement.

Kevin Weil

I totally agree. It's crazy exciting.

Peter Diamandis

Is that a fertile area for startups, or should they stay away from that because it's just where the big boys play?

Kevin Weil

I totally think it's a fertile area. It takes a lot of technical talent to do it, but I think that's an interesting spot. I know some startups doing exciting stuff there—material science and others.

Peter Diamandis

Yeah. Material science is an unsung hero, and it will be a huge unlock.

Kevin Weil

Yeah.

Peter Diamandis

Oh, it'd be amazing. Yeah.

And Lieutenant Colonel Weil.

Kevin Weil

Yes.

Peter Diamandis

June 13th, you're inducted into the Army.

Kevin Weil

Yeah. Super excited about it.

Peter Diamandis

Super excited. Did you go through Army basic training?

Kevin Weil

We did a version of basic training. It was kind of an accelerated version.

Peter Diamandis

I mean, you're an ultramarathoner, so I mean—

Kevin Weil

Probably not. But, you know, we went, and we've all passed our Army fitness tests.

Peter Diamandis

So, tell us, how did that materialize? Did you get a call one day?

Kevin Weil

Yeah. I got a call from Shyam Sankar, who is the CTO of Palantir. He's been there forever. He and some of the folks at the top of the Army were thinking about this program.

Peter Diamandis

I mean, it's really smart. I was impressed.

Kevin Weil

Yeah. Kudos to the Army for being willing to take a risk on some crazy people like us.

Peter Diamandis

But I think it's awesome. By the way, I hope it's 4 of us today.

Kevin Weil

Yes. It's myself, Shyam, Boz, who's the CTO of Meta, and Bob McGrew, who used to be the head of research here. The idea is to bring tech and the military—the DoD—closer together.

Peter Diamandis

Take a bunch of expertise that you get from working in the tech industry, a bunch of expertise that you get from being in the DoD, and merge them together, because we'll be better together. I imagine the Navy and the Air Force need to do something very similar.

Kevin Weil

Yeah. There's a huge push across the DoD to integrate AI, as there very well should be. It does us no good to have the best models in the world if they're sitting on the shelf—

Peter Diamandis

Yeah, while the PLA uses inferior models but integrates them everywhere.

Kevin Weil

Yeah, no doubt. So it's super important. The idea here was to bring AI across the DoD.

Peter Diamandis

There are lots of efforts to bring AI across the DoD, as there very well should be. Palmer Luckey is a friend, and he's been doing an amazing job at Anduril.

Kevin Weil

Oh, totally. Thank God for them.

Peter Diamandis

But you could have done this as a consultant. I think getting us in on the inside—first of all, wearing the uniform, you feel a responsibility. You go hang out with people who are literally giving their lives, but also, it's such an incredible institution that you wouldn't get that sense on the outside. I didn't have that sense before. I respected it, obviously, but being on the inside is a different thing.

Kevin Weil

Obviously, I have a ton to learn. We're just getting started. But I think us being on the inside will allow us to be more effective. When we go wherever they'd like us to go to help, we'll have that much more of a sense of how everything works, and we'll be part of the team, as opposed to being a consultant from the outside.

Peter Diamandis

Yeah. So, you have to go through the whole top-secret clearance process and all the background checks and all that. Was that—

Kevin Weil

I had one already.

Peter Diamandis

Did all 4 of you have it already?

Kevin Weil

Not all of us, but they had us go through it.

Peter Diamandis

So then, once you're inside, on base or whatever, do you just have access to everything? How does that work?

Kevin Weil

I mean, it's still very—top-secret stuff is pretty heavily compartmented. There's a need-to-know basis for most things.

Peter Diamandis

Yeah.

Kevin Weil

So they certainly don't just open up everything. But where there are things that are important for us to know, we can know, and we can help in deeper ways than if we didn't have that access.

Peter Diamandis

Is there a schedule where you go once in a while?

Kevin Weil

It's a little more ad hoc than normal reserve duty. Normal reserve duty is kind of a weekend a month and 2 weeks in the summer. This is much more—because of the jobs, it's harder to be a weekend a month, but we can also go maybe deeper and longer for specific things. We're also each going to take our own focus.

Peter Diamandis

You obviously have to watch out for conflicts, and they've designed your focus areas based on that. What's your focus?

Kevin Weil

One of the things I'm going to focus on is physical performance and how you can use AI to monitor and improve physical performance.

Peter Diamandis

For sure.

Kevin Weil

You know, I've got a Whoop, an Oura, and an Apple Watch.

Peter Diamandis

By the way, which gives you better sleep data, do you think—Whoop or Oura?

Kevin Weil

They disagree, and I never quite know.

Peter Diamandis

I know. I've got an Eight Sleep as well. I got one of my best night's sleep last night. I got a 94 score.

Kevin Weil

Yes, that was great.

Peter Diamandis

That's amazing. The one thing mine agree on is that I need to sleep more and get more deep sleep.

Kevin Weil

Yeah.

Peter Diamandis

Yeah. Let's wrap with advice for entrepreneurs right now. There are a lot of entrepreneurs and builders here. As a chief product officer building products, what sage advice do you have for them?

Kevin Weil

Oh, man. I think lean into AI in every way possible, and assume that it will continue on the super-sharp, steep curve that it's on. Build—just imagine things that should be true and are not possible today. I think a lot of them are going to be possible to build 6 months from now, a year from now, or 2 years from now. The people who see that and are building for that future, counting on the models to get there—that's going to pay off, because the models are going to get there.

Peter Diamandis

Yeah. All right. I do have an important last question for you. Has OpenAI fully reached AGI, and what is Jony Ive delivering for you?

Kevin Weil

Oh, man. I've been waiting to talk about this since we started.

Peter Diamandis

Really? Okay.

Peter Diamandis

All right. So, [Laughter] and that's a cut, ladies and gentlemen. So, that's a wrap. Obviously, we didn't get information from Kevin that's exclusive on Jony Ive or AGI as much as I would love to, but hopefully, if you're a builder and entrepreneur at home, this was super useful for you. I know it was for me.

Open AI Insider on GPT-5, AGI & the Great AI Race w/ Kevin Weil & Dave Blundin | BidClub