[0:00:05] Hello, I'm Sam. This is our chief scientist, Jakob, and we have a bunch of updates to share today about OpenAI. Obviously, the news of today is our new structure. We're going to get to that near the end, but there's a lot of other important context we would like to share first. Given the importance of a lot of this, we're going to go into an unusual level of transparency about some of our specific research goals and infrastructure plans and product, but we think it's sort of very much in the public interest at this point to cover all of this. So, our mission at OpenAI...
[0:00:33] in both the nonprofit and our new PBC is to ensure that artificial general intelligence benefits all of humanity. As we get closer to building this, we have new insights into what that is going to mean. There was a time earlier on in OpenAI where we thought that AGI would be sort of this oracular thing in the sky and it would make all these wonderful things for us. And we now have a sharper view of that, which is we want to create tools.
[0:00:58] And then we want people to use them to create the future. We want to empower people with AI as much as possible and then trust that the process that has been working for human history of people building better and better things with newer and better tools will continue to go on. We can now see a vision where we help build a personal AGI that people can use anywhere with all of these different tools, access to all these different services and systems to help with work and personal life and their personal life.
[0:01:26] AI gets better and better as AI can even do things like discover or help discover new science. What people will be able to create with that to make all of society better and their own lives more fulfilled, we think should be quite incredible. There are three core pillars we think about for OpenAI, research, product, and infrastructure. We have to succeed at the research required to build AGI. We have to build a platform that makes it easy and powerful to use. And we have to build enough infrastructure such that people can use at a...
[0:01:54] Low cost, all of this amazing AI that they like. Here's a little cartoon of how we think about our world. So at the bottom layer here, we have chips, racks, and the systems around them, the data centers that these go into, and the energy. We'll talk more about the first three today in energy another time. Then we train models on top of these. Then we have an OpenAI account on top of that. We have a browser now called Atlas, and we have devices coming in the next few years that you'll be able to take AI with you everywhere.
[0:02:23] And we have a few first-party apps like ChatGPT and Sora, and we'll have more over time. But mostly what we're excited about is this big puzzle piece in the upper right. We're finally getting to a world where we can see that people are going to be able to build incredible services with AI, starting with our API, with apps in ChatGPT, new enterprise platform that we'll have over time, an OpenAI account, and way, way more. And people will be able to fit.
[0:02:46] All of the kind of like current things in the world and many more into this new AI world. And we want to enable that. And we believe the world will build just a huge amount of value for all of us. So that's kind of what we see the economic picture looking like. But one of the things that we've thought about for a long time and we really see happening now or starting to happen now, glimmers of it, green shoots, whatever you want to call it, is the impact that AI will have on science. And although the economic impact from that previous slide would be huge, for the long-term quality of life,
[0:03:16] And improvement and change in society. AI that can autonomously discover new science or help people discover new science faster will be, I think, one of the most important things and something that we're really trying to wrap our heads around. So I'm going to hand this over to Jakob to talk about research. And as I mentioned, we're going to share a lot about our internal goals and our picture of where things are. Thanks, Ham. At Dakar, we are a research laboratory.
[0:03:45] focused on understanding a technology called deep learning. And so a particular focus of ours is understanding what happens as you scale up training deep learning systems. And one consequence we discuss a lot there is AGI, artificial general intelligence. But we find that in some way, even this maybe understates a bit.
[0:04:12] The magnitude of the possible progress and change here. And so in particular, we believe that it is possible that deep learning systems are less than a decade away from super intelligence. So systems that are smarter than all of us on a large number of critical axes. And this is, of course,
[0:04:45] A serious thing, right? There is a lot of implications of this to grapple with. And one particular focusing impact of this technology and the technologies leading up to it and something that we organize our entire research program around is the potential to accelerate scientific discovery, to accelerate the development of new technology. We believe that this will be
[0:05:15] Perhaps the most significant long-term impact of AI development. And it will fundamentally change the pace of progress on developing new technologies. And so thinking about how far along we are towards these goals, one good way to think about progress is to look at the time horizon that it would take people to accomplish the tasks that the models can perform.
[0:05:46] And so this is something that has been extending rapidly over the past few years. So where the current generation of models is at right now is about five hours. So you can see this by looking at the models matching the best people in competitions such as the International Olympiad in mathematics or informatics. And we believe that this horizon will continue to extend rapidly. And this is in part as a result of algorithmic innovation and a part
[0:06:16] Just scaling deep learning further. And in particular, scaling along this new axis in context compute, also called test time compute, where we really see orders and orders of magnitude to go. So this is roughly how much time the model spends thinking. And if you look at how much time the models currently spend thinking about problems, and if you think about how much
[0:06:40] Compute how much time you would like to spend on problems that really matter, such as scientific breakthroughs. You should be OK using entire data centers. And so there is really quite a way to go there. And anticipating this progress, we, of course, make plans around it internally. And we want to provide some transparency around our thinking there. And so we want to take this maybe somewhat unusual step of sharing our internal goals and goal timelines towards these very powerful systems.
[0:07:11] And these particular dates, we absolutely may be quite drunk about them. But this is how we currently think. This is currently how we plan and organize. And so as a research organization that is working on automating research, naturally, we are thinking about how does this impact our own work? And how will AI systems that accelerate development of future AI systems look like? How can they empower research like alignment?
[0:07:40] We are making plans around getting to quite capable AI research interns that can meaningfully accelerate our researchers by expanding a significant amount of compute by September of next year. So we believe that it's actually quite close. And then we look towards getting a system capable of autonomously delivering on larger research projects and a meaningful, fully automated AI researcher
[0:08:10] by March of 2028. And so, of course, as we look towards these very capable systems, we think a lot about safety and alignment. And in fact, a lot of our work, both on deployments and safety, but also just on understanding deep learning and the development capability side, we can think of preparation for these very capable models. Safety is a multifaceted problem.
[0:08:39] And so the way we generally structure our thinking are these five layers ranging from factors that are most internal to the model to ones that are most external. And so at the core, right, and what we believe is the most important long-term safety question for superintelligence is value alignment. So to put this, value alignment you can think of as
[0:09:10] What is really the thing that the AI fundamentally cares about? Can it adhere to some high-level principles? What will it do if it's given unclear and conflicting objectives? Does it love humanity? And the reason we believe that this high-level objectives or principles driving the AI are so important is that
[0:09:41] As we get to the systems that are thinking for very long, as they become very smart, as they tackle problems that are at the edge or perhaps beyond human ability, really getting to complete specifications becomes quite difficult. And so we have to rely on this deeper alignment. Then there's goal alignment. Does the agent
[0:10:08] Interact with people. How does it interact with people? How does it do it following instructions? Then reliability. Can the AI correctly calibrate its predictions? Can it be reliable on easy tasks, express uncertainty on hard ones? Can it deal with environments that are a little bit unfamiliar? Then we have adversarial robustness, which is very related to reliability, but it's about adversarial settings. So can the AI withstand
[0:10:37] Targeted attacks from human or AI adversaries. And then the outer layer is systemic safety, which are guarantees about the behavior of the overall system that do not rely on the AI's intelligence or alignment. So for example, this can be security, or what data does the AI have access to, or what devices it can use. And so we invest in multiple research directions across these domains.
[0:11:07] And we have seen quite a lot of progress also come from just the general development and improving understanding of deep learning as a whole. And I want to take a slightly deeper technical dive here and talk about the particular direction. Value alignment is a hard problem, right? It's definitely not solved yet. However, there is a new promising tool that aids our study of it.
[0:11:37] And that is chain of thought faithfulness. It's something we invest in very heavily. Starting from our first reasoning models, we have been pursuing this new direction in interpretability. And the idea is to keep parts of the model's internal reasoning free from supervision. So don't look at it during training. And thus let it remain representative of the model's internal process. So we...
[0:12:07] Refrain from kind of guiding the model to think good thoughts and so let it remain a bit more faithful to what it actually thinks. And this is not guaranteed to work, of course. We cannot make mathematical proofs about deep learning. And so this is something we study. But there are two reasons to be optimistic. One reason is that we have seen very promising empirical results. This is a technology we employed a lot.
[0:12:37] Internally, we use this to understand how our models train, how their propensities evolve over training. Also, we have had successful external collaborations on investigating the models' propensity to scheme, for example. And secondly, it is scalable.
[0:13:01] And in the sense that explicitly, we make the scalable objective not adversarial to our ability to monitor the model. And of course, an objective not being adversarial to the ability to monitor a model is only half the battle. And ideally, you want to get it to help with monitoring the model. And so this is something we're researching quite heavily. But one important thing to underscore about chain of thought faithfulness is it's somewhat fragile.
[0:13:32] It really requires drawing this clean boundary and having this clear abstraction and having restraint in what ways you can access the chain of thought. And this is something that is present at OpenAI from algorithm design to the way we design our products. So if you look at the chain of thought summaries in ChatGPT, if we didn't have the chain of thought summarizer, if we just
[0:13:56] Make the chain of thought fully visible at all times. That would make it kind of part of the overall experience. Over time, it would be very difficult to not subject it to any supervision. And so long term, we believe that by preserving some amount of this controlled privacy for the models, we can retain the ability to understand their inner process. And we believe this can be a very powerful technique as we move towards these very capable, long-running systems. And I'll hand back to Sam.
[0:14:24] Okay, that's very hard to follow with the rest of this. And obviously, that's the most important part of what we have to say. But, you know, just to reiterate, we may be totally wrong. We have set goals and missed them miserably before. But with the picture we see, we think it is plausible that by September of next year, we have sort of an intern level AI research assistant. And that by March of 2028, which I believe is almost five years to the month after the launch of GPT-4.
[0:14:52] We have a legitimate AI researcher, and this is the core thrust of our research program. There are two other areas we want to talk about, product and then infrastructure. On the product side, as we make this incredible progress with deep learning, we want to make it useful to people to sort of invent the future, as we mentioned. And what that's looked like traditionally for us is an AI super assistant inside of ChatGPT, but we're now really going to evolve to a platform that other people will build on top of and all of the pieces that need to fit in of the world.
[0:15:22] You know, will be built by others. Before we go talk about that, I wanted to just show a quick video of how people are using GPT-5, some of the ways people are using GPT-5 in ChatGPT today. I'm a quantum physicist. I'm a nail technician. Human immunologist. I'm a steel worker. A designer and developer. A professor of economics. I basically go fishing for a living.
[0:15:44] GPT-5 is able to predict the outcomes of experiments that we haven't even done. Can you create a part numbering system that's easy for my guys in the shop? I want to catch Dungeness crab.
[0:15:59] You would ask a question about the application of a certain quantum operator. This one gives me a very detailed mathematics. I ask for basically a camera app where I can draw real time in the air. We have a theme, we have a direction to go.
[0:16:16] Let's go from a million or infinite amount of ideas to, like, give me 10 or 20. You just start testing it. You know, you tell it jokes. You ask it questions like, what should economists do? Different baits I can use, different water depths. All this information that it would take years to figure out under a lot of trial and error. It's tremendous for brainstorming. It's back and forth. I can kind of easily follow.
[0:16:42] Jacob will join me back for Q&A in a little bit, but we're going to have one special guest on before the end. We love that. We want much more of that. We want that everywhere. So we want OpenAI to be a platform that people...
[0:17:13] And companies can build on top of. We can sort of see our way now to an AI cloud where this is not just in ChatGPD. This is not just services that we create. But we want to expose our technology for as many people to build the things that people will depend on and use and create with as possible. I think this quote, or at least this idea, is originally from Bill Gates. At least that's where I first heard it. That you know you've built a platform when there's more value created by people building on the platform than by the platform builder. And that's our goal.
[0:17:43] And that's our goal next year. We really think we can now take this technology and this user base and this sort of framework we've built and get the whole world to build amazing new companies and services and applications on top of it. To do that, there will be many things that we have to evolve towards, but there's two foundational principles as we move towards being this platform that I wanted to touch on. One is about user freedom. If this is going to be a platform that all sorts of people are building on using, creating with,
[0:18:14] People around the world have very different needs and desires. And there will, of course, be some very broad bounds. But we want users to have a lot of control and customization of how they use it. Now, I made one of my many stupid mistakes when I tried to talk about this recently. I wish I had used an example other than erotica. I thought there was an understandable difference between erotica and porn bots. But in any case, we were trying to show the point we're trying to get across.
[0:18:39] is that people need a lot of flexibility and people want to use these things in different ways. And we want to treat our adult users like adults. In our own first party services, we may have tighter guidelines, but AI is going to become such an important part of people's lives. The freedom of human expression is going to need to be there. Along with that, we think that world will need to think about privacy in a different way than they have for previous kinds of technology. Privacy is important for all sorts of technology, of course, but privacy for AI
[0:19:07] will be especially important. People are using this technology in a different way than they've used the technologies of the past. They're talking to it like they would to their doctor or their lawyer, their spouse. They're sharing the most intimate details of their lives. And of course, we need strong technical protections on that privacy, of that privacy. But we also think we need strong policy protections of that privacy. We've talked about concepts like AI privilege, but really strong protections if AI is going to be this fundamental platform in people's lives.
[0:19:35] Okay, and then I want to go on to infrastructure. So I know there's been a lot of confusion about sort of where we are in our infrastructure build-out, and we figured we would just be super transparent about that. So where we are today, all of our commitments total a little bit over 30 gigawatts of infrastructure build-out, and that's about a $1.4 trillion total financial obligation for us over the next many years. This is...
[0:20:04] What we've committed to so far, we, of course, hope to do much more. But given the picture we see today, given what we think we can see for revenue growth, our ability to raise capital, this is what we're currently comfortable with. This requires a ton of partnerships. We've talked about many of our great chip partners. There are people building the data centers for us, land, energy. There will be chip fab facilities. This is already getting to require quite a lot of supply chain innovation. And we're thrilled to get to work with.
[0:20:30] AMD, Broadcom, Google, Microsoft, Nvidia, Oracle, SoftBank, many others to really make this happen. But this is still early. If the work that Jakob talks about comes to fruition, which we think it will, and if the economic value of these things happen and people want to use all these services, we're going to need much more than this. So I want to be clear. We're not committing to this yet, but we are having conversations about it. Our aspiration is that we can build an infrastructure factory.
[0:20:57] Where we can create one gigawatt a week of compute and we aspirationally would like to get that cost down significantly to like 20 billion dollars a gigawatt over the five-year life cycle of that equipment. To do this will require a ton of innovation, a ton of partnerships, obviously a lot of revenue growth. We'll have to repurpose our thoughts about robotics to help us build data centers instead of doing all the other things. But this is where we'd like to go.
[0:21:24] And over the coming months, we are going to do a lot of work to see if we can get here. It will be some time before we're in a financial position where we could actually pull the trigger and get going on this. One gigawatt is like a big number, but I figured we would show a little video to put this into perspective. This is a data center that we're building in Abilene, Texas. This is the first Stargate site. We're doing several of these now around the country, but this one is the furthest along. There's like many thousands of people that work here every day just doing the construction at the site.
[0:21:54] There's probably hundreds of thousands or millions of people that work in the supply chain to make all this happen, to design these chips, to fab these chips, to put them together. There's all of the work that goes into this for energy. There's an enormous amount of stuff that has to happen for each one gigawatt. And we want to figure out how we can make this way more efficient and way cheaper and way more scalable so that we can deliver on the infrastructure that the research roadmap requires and that all of the ways that people will want to use this need.
[0:22:23] To enable that, we have a new structure. So maybe you saw before this like crazy convoluted diagram of all of the OpenAI entities. Now it's much simpler. We have a nonprofit called the OpenAI Foundation that is in control of where the board sits, or where, come back to the board, where the board also sits, and owns a slice of our PBC, Public Benefit Corporation, called OpenAI Group.
[0:22:51] So nonprofit in control, public benefit corporation sits under it. We hope for the OpenAI Foundation to be the biggest nonprofit ever. As I mentioned now a few times, science is one of the ways that we think the world most improves, along with the institutions that broadly distribute the benefits of that. So then science will not be the only thing that the nonprofit funds, but it will be an important first major area of the things that we do. The nonprofit will govern the PBC.
[0:23:22] It will initially own about 26% of the PBC equity, but that can increase over time with warrants if we perform really well. And it will use these resources to pursue what we think are the best benefits of AI, given where the technology is and what society needs. The PBC will operate more like a normal company. It will have the same mission. It will be bound to that mission. And matters of safety will only be bound to that mission.
[0:23:48] But it will be able to attract the resources that we need for that gigantic infrastructure build out to serve the research and product goals that we have. So the initial focus of the foundation, we'll do more things over time, but we want to knock something out of the park hopefully first, is a $25 billion commitment to use AI to help cure disease. There are a lot of ways this can happen, generating data, using a lot of compute grants to scientists, and also for AI resilience.
[0:24:17] AI resilience is a new and I think very important area. And I'd like to invite our co-founder, Wojciech, up to talk about what this will look like. Hello. Glad to be here. Thanks for being here. So term AI resilience is a little bit broader than what we historically thought about AI safety. So in case of the resilience, we think that
[0:24:43] Advanced AI comes with risks and disruptions. And we would like to have an ecosystem of organizations that can help to solve a number of these problems. So let me give you an example to better illustrate it. We all believe that AI
[0:25:03] will advance in biology and as it advances in biology there is a risk that some bad actor could use AI to create man-made pandemics. So on the safety level the mitigation would be to make sure that the models block the queries that have to do with virology.
[0:25:27] However, if you consider the entire AI industry, it's very likely that even if OpenAI blocks it, someone could use different models out there and still produce the pathogens.
[0:25:43] And in case of resilience, we don't want just to block it, but also have a rapid response if the problem would occur. So when I think about the risks and disruptions, there are just many.
[0:25:58] Mental health is one of them. Bio is another one. Job displacement might be another one. And we think that we need the ecosystem. Maybe a good analogy that I like is cybersecurity. So at the beginning of internet, it was actually, it was a place that people didn't feel comfortable putting their credit card numbers because it was so easy to get hacked.
[0:26:24] And when there was a virus, people are giving each other a call to disconnect the computer from internet. And we got there a long way. At the moment, there's entire infrastructure of cybersecurity companies. They are protecting the critical infrastructure, governments, corporations, and individual users to such extent that people are willing to put the most personal data.
[0:26:51] So am I. I think this is an important time to be doing this and I'm very excited.
[0:27:20] You're going to figure out how we go off and make it happen. So again, these are not the only things that the nonprofit will fund, but we're excited about these as the first two, using AI to develop cures and treatments for diseases and this new AI resilience effort as we figure out what the deployment of AGI into society is going to look like. So we mentioned that those are our three pillars, but what if this all works? We think it is plausible that in
[0:27:49] 2026, we start to see the models of that year begin to make small discoveries. By 2028, medium, or maybe even larger discoveries. And who knows what 2030 and 2032 are going to look like. If AI can keep advancing science, as has happened in the past, we think the future can be very bright.
[0:28:09] Of course, we think it's very important that humans can self-determine our way through this future. But the open space that new scientific advances give us is quite impressive. So we asked Sora to help us imagine a radically better future by looking at the past. And we are particularly interested in how the history of science builds on itself, discovery after discovery. This is what we hope will happen with AI. You know, this is going to be 200 years of science.
[0:28:35] But if you can do these 200 years of compounding discoveries of the scaffolding building up on each other, not in 200 years, but in 20 years or in two years, and if you look at how much this is accelerated, think about what could be possible before, you can imagine a world where a radically better future becomes quite possible. You have a data center here that is discovering a cure for cancer, a data center there that's making the best entertainment ever, a data center here that's helping you find your future.
[0:29:02] Husband or wife, this one is building rockets and helping you colonize space. This one is helping to solve the climate crisis. So we did all this stuff the old fashioned way. And now with the help of AI, we'll be able to shape what comes next with maybe much more power. So we talked a little about AI medicine. We're very excited about robots. We really think energy is very, very important to the world. We want to figure out what personalized education can mean, design novel materials, and probably a ton of other things that we can't even think of yet.
[0:29:31] As we head into this next phase of OpenAI, and more importantly than that, this continual progress in deep learning, we thank you for joining us today. And we're going to try something new now, which is we're going to just answer questions. If this works, it's something we'll try more in the future. Jakob is going to rejoin for this Q&A. Thank you very much, Wojciech. Thank you, Sam. But this is a new format for us, so bear with us as we try it this first time. Again, if this is useful, it's something we'll do again a lot more.
[0:29:58] And we're going to try to just answer questions in the order they are most upvoted. Are we good to go? All right, let's see how this works. So you can put questions in the Vimeo link, and we will just start answering them. So from Caleb, we've warned that the tech is becoming addictive and eroding. The tech is becoming addictive and eroding trust, yet Sora mimics TikTok and ChatGPT may add ads. Why repeat the same patterns you criticized, and how will you rebuild trust through actions and not just words? We're definitely worried about this. I worry about it.
[0:30:29] Not just for things like Sora and TikTok and ads and ChatGPT, which are maybe known problems that we can design carefully. But, you know, we have certainly seen people develop relationships with chatbots that we didn't expect. And there can clearly be addictive behavior there. Given the dynamics and competition in the world, I suspect some companies will offer very addictive new kinds of products. And I think you'll have to just judge us on our actions. We'll have to, you know...
[0:30:56] We'll make some mistakes. We'll try to roll back models that are problematic. If we ship SOAR and it becomes super addictive and not about creation, we'll cancel the product, and you'll have to just judge us on that. My hope and belief is that we will not make the same mistakes that companies before us have made. I don't think they meant to make them either. We're all kind of discovering this together. We probably will make new ones, though, and we'll just have to evolve quickly and have a tight feedback loop. We can imagine all sorts of ways this technology does incredible good in the world.
[0:31:25] Also obvious bad ones. And we're guided by a mission where we'll just continuously evolve the product. One thing that we are quite hopeful about in terms of what we optimize for in products like ChatGPT or Sora is thinking about optimizing for the very long term, which is naturally very aligned with how we think in general about extending the horizon on which the models can work productively.
[0:31:56] And so we believe that quite a lot of development is possible there, and we can eventually get models that really optimize for long-term satisfaction and well-being instead of just short-term signals. Okay, next question. Will we have an option to keep the 4.0 model permanently after adult mode is installed? We don't need to save the models to responsible adults. We have no plans to sunset 4.0.
[0:32:27] We are not going to promise to keep it around to the heat death of the universe either, but we understand that it's a product that some of our users really love. We also hope other people understand why it was not a model that we thought was healthy for miners to be using. We hope that we build better models over time that people like more. The people you have a relationship with in your life, they evolve and get smarter and change a little bit over time. We hope that the same thing will happen.
[0:32:54] But yeah, no plans to sunset 4.0 currently. Wow, we have a lot of 4.0 questions. All right, in the interest of time, we will not go through all of these. But yeah, we want people to have models that they want to use. We don't want people to feel like we're routing them around models. And we want adults to make choices as adults, as long as we think we're not selling heroin or whatever, which also you shouldn't do.
[0:33:25] People that want to have emotional speech, as we've said, we want to allow more of that, and we plan to. OK, here's a good anonymous question for Jakob. When will AGI happen? So I think in some number of years, we'll look back at these years, and we'll say this was kind of the transition period when AGI happened. I think one way we thought about
[0:33:58] I think, as Sam said, early on at OpenAI, we thought about AGI emotionally as this thing that is the ultimate solution of all the problems. And it's this single point for which there is before and after. And I think we found that it's a bit more continuous than that. And so in particular, for various kind of
[0:34:27] Benchmarks that seemed like the obvious milestones towards AGI, I think we now think of them as indicating roughly how far away we are in years. And so if you look at the succession of milestones such as computers beating humans at chess, and then at Go, and then computers being able to speak in natural language, and computers being able to solve
[0:34:56] Math problems, right? I think, well, they clearly kind of get colored together. Yeah, I would say I think it's the AGI term has become hugely overloaded. And as Jakob said, it'll be this process over a number of years that we're in the middle of. But one of the reasons we wanted to present what we did today is I think it's much more useful to say our intention, our goal is by March of 2028 to have a true automated AI researcher.
[0:35:26] And define what that means than it is to sort of try to satisfy everyone with a definition of AGI. And maybe one other thing to mention. I think one kind of counterintuitive thing here is that obviously we're working with a pretty complicated technology. We're trying to understand all these algorithms. And maybe initially we kind of imagined that AGI is the moment where you kind of have figured out all the answers. And it's kind of the final thing. And I think now we increasingly realize that there is kind of a
[0:35:56] Can you give us an idea of how far ahead internal models are compared to deployed ones?
[0:36:29] I think we have quite strong expectations for our next models. So I think we expect quite rapid progress over the next couple of months and a year. Yeah, I think we haven't been withholding something.
[0:37:00] Extremely crazy. One of the ways this kind of often works in practice is there's like a lot of pieces that we develop and that, you know, they're all kind of hard won victories. And then we know that when we put them together, we will have something quite impressive. And we're able to predict that fairly well. Part of our goal today is to say that we have a lot of those pieces. It's not like we're kind of currently sitting on this giant thing that we're not showing to the world, but that we expect.
[0:37:28] By a year from now, certainly with this September of 2026 goal, that we have a realistic shot at a tremendously important step forward in capability. Ronan asks, what is OpenAI's stance on partnering with labs like Anthropic, Gemini, or XAI for joint research, compute sharing, and safety efforts? We think this is going to be increasingly important on the safety front. Labs will need to share safety techniques, safety standards.
[0:38:00] You can imagine a time when the whole world would say, okay, before we hit a recursive self-improvement phase, we really need to all carefully study this together. We welcome that collaboration. I think it'll be quite important. One thing to mention on the chain of thought faithfulness that I talked about earlier, we actually have started talking about establishing industry in arms and we started some joint investigations with researchers from Google and Anthropic and some other labs.
[0:38:30] Yeah, that's something I'm very excited about, and I think that is an example of something where we can really benefit from collaborating across multiple labs. Anonymous asks, will you ever open source some of your old models, like the original GPT-4? We might do those as museum artifacts someday, but they're not, like, GPT-4 is not a particularly useful open source model. It's big. It's not that good. You know, we could probably make something that is beyond the power of GPT-4 at a very tiny scale that actually would be useful to people. So for useful things,
[0:38:59] I expect more things like that. For fun museum artifacts, yeah, someday. Who knows? I think there could be a lot of cool things like that. Another anonymous, or maybe the same one, asks, will you admit that your new model is inferior to the previous one and that you're ruining your company with your arrogance and greed while ignoring users' needs? I believe that it is inferior to you for your use case, and we would like to build models that are better for your use case. On the whole, we think for most users, it's a better and more capable model.
[0:39:28] We definitely have learned things about the 4.0 to 5.0 upgrade, and we will try to do much better in the future, both about better continuity and about making sure that our model gets better for most users, not just sort of people that are using AI for science or coding or for whatever. Yum asks, will there ever be a version of ChatGPT meant for personal connection and reflection, not only business or education? Yeah, for sure. We think this is a wonderful use of AI.
[0:39:56] We're very touched by how much this has meant to people's lives. All of us get a ton of emails and outreach from users about how ChatGPT has helped people in difficult personal situations or to live a better life. And this is what we're here for. I mean, this is as important as anything that we do. We love to hear about scientific progress. We love to hear about people that got diagnosed with the disease and got cured. But the personal stories are incredibly important to us. And we're thrilled about that. And we absolutely want to offer.
[0:40:25] Such a service. Okay, two questions that are about Tide from G and Anonymous. Your safety routing breaks user trust and workflows by overriding our choices. Will you commit to revoking this paternalistic policy for all consenting adult users and stop treating us like children? When do users get control over routing? Where's the transparency on safety and censorship? Why can't adults pick their own models? Yeah, I don't think the way we handled the model routing was our best.
[0:40:55] There are some real problems with 4.0, and we have seen a problem where people that are in fragile psychiatric situations using a model like 4.0 can get into a worse one. Most adult users can use those fine, but we do, as we've mentioned, we have an obligation to protect minor users, and we also have an obligation to protect adult users who are not in a frame of mind where we're reasonably likely that they're choosing what they really want and we're not causing them harm.
[0:41:24] As we build age verification in and as we are able to differentiate users that are having a true mental health crisis from users who are not, we of course want to give people more user freedom. As we mentioned, that's one of our platform principles. So yes, expect improvement there. And I don't think this was our best work in how we communicated the previous rollout. How we strike the right balance between protecting people and allowing adults to speak about difficult things without feeling policed.
[0:41:55] You want to say anything there? Yeah. So definitely there is a problem where we aim to lay out the kind of high-level policies and guidelines for the model in the spec that we develop for ChartGPT.
[0:42:29] The space of situations you can find yourself in is enormous. And at some point, establishing the right boundaries really becomes a tough intelligence problem. And so we are seeing improved results on this metrics from reasoning models and from expanding more.
[0:42:57] More reasoning on thinking about this software questions and trade-offs. Of course, this is a bit more difficult to train for also than math problems, for example. And so this is something that we're researching quite heavily. Kate says, when in December will adult mode come? Will it have more than just NSFW when writing even slight conflict when writing triggers filters on Foro?
[0:43:26] I don't know exactly when in December it will ship, but yes, the goal is when you are writing, when you are using OpenAI to help ChatGPT to help you with creative writing, it should be much more permissive in many categories than the previous models are. Again, we want this, and we know users want this too. If this is going to be your personal tool, it should help you with what you're doing, and every time you hit a content filter for something that feels like...
[0:43:53] It shouldn't. We understand how annoying that is. So we're going to try to fix that with adult mode. There may be new problems that we face, but we want to give people more flexibility. Anonymous says, why does your idea of safety require lying to users about what model they're actually using? Again, I think we misrolled this one out, but the goal here was to let people continue to use 4.0, but in the situations where 4.0 has behavior that we think is actually really harmful before we have all of the age-gaming that we'd like.
[0:44:23] To put the user into a model where they are not going to have some of the mental health problems that we faced with 4.0. 4.0 was an interesting challenge. It's a model that some users really love, and it was a model that was causing some users harm that they really didn't want. And I don't think this is the last time we'll face challenges like this with a model. But we are trying to figure out the right way to balance that. Will we be getting the legacy models back for adults without rewriting? Yes.
[0:44:54] Will the December update officially clarify OpenAI's position on human-AI emotional bonds, or will restrictions continue implicitly defining such connections as harmful worldwide? I don't know what it means to have an official position. We build this tool. You can use it the way you want. If you want to have a small R relationship and you're getting something like empathy or friendship that matters to you and your life out of a model.
[0:45:21] It's very important to us that the model faithfully communicate what it is and what it isn't. But if you as the user are finding value in that support, again, we think that's awesome. We are very touched by the stories of people who find value, utility, better life in the emotional support or other kinds of support they get from these models. Kylo says, how is OpenAI increasingly allowing so many features for the free version users?
[0:45:48] I can answer this from a product and business perspective, but Jakob, I think it might be useful for you to just talk about the incredible rate at which models are getting more capable for lower prices and less amounts of compute. Yeah, we are seeing quite a lot of ability as we get to the new frontiers of intelligence to reduce the cost for that quite quickly. And so, yeah, especially with reasoning models, we've seen
[0:46:20] That actually quite cheap models, when allowed some additional test time compute, can become much more capable. And so this is something that I expect will continue. And so as we talk about getting to these new frontiers and automating research and so forth, I expect that.
[0:46:48] The cost of all of that will keep falling quite a lot too. So yeah, we talk a lot about the increase in model capability and for pushing forward science, that's hugely important. One of the most amazing things that I've observed about AI is over the last few years, the sort of price of a particular unit of intelligence has fallen about 40x per year for the last few years.
[0:47:19] So when we first had GPT-3, we thought it was very cool. And it was at this cost, it was kind of hard. And GPT-3 scale models now basically run for free on a phone or something. The cost of a model that's as smart as GPT-4 at the time we launched it relative now has fallen hugely. And we expect this trend to keep going. Now, we still think we need a ton of infrastructure because what we continue to find is the cheaper we can make it, the more people want to use it. And I expect that only to increase.
[0:47:49] But our goal is to drive the cost of intelligence down and down and down and have people use it for more and more things. That will allow us to continue to offer lots of features for free. But that will also mean, I think, that people who really want to spend a lot on pushing the art to the limit to cure a disease or figure out how to build a better rocket or whatever will spend a huge amount. We are committed to continuing to put the best...
[0:48:15] Technology we can, as long as we can make the business model even sort of work into the free tier. And you should expect a lot more from us there over time. Okay, Anonymous asks, will an age verification start that allows users to opt out of the safety route or a waiver that could be signed releasing OpenAI from any liability? We're not gonna like do, again, we're not gonna do the equivalent of selling heroin or whatever, even if you sign a liability, but yes.
[0:48:44] On the principle of treat adult users like adults, if you're age verified, you will get quite a lot of flexibility. We think that's important. And clearly, it resonates with the people asking these questions. Anonymous also asks, is ChatGPT the Ask Jeeves of AI? We sure hope not. We don't think it will be. OK. Since we only have 10 minutes left, we're going to, and some of these touch other things that we've
[0:49:12] We've already touched on. We're going to skip down through some of the same questions and try to get to more. In future sessions, we can do more of these if we don't get to everything here. Noah asks, just how the Macintosh was the precursor to the iPhone. Do you see ChatGPT as the OpenAI product, or do you see it as a precursor to something much greater that truly reshapes the world? So I would say, as a research lab, we haven't set out to build
[0:49:47] A chatbot originally, although I think we've since come to appreciate how aligned this product is with our overall mission. And we of course expect ChatGPT to continue to become better and be this way for people to interact with increasingly advanced AI. But we do anticipate that eventually AI systems will be capable of creating valuable artifacts of
[0:50:17] Actually pushing scientific progress forward as we were discussing. And I believe that will be the real lasting legacy of AI. I think the chat interface is a great interface. It won't be the only interface. But the way that people use these systems will change hugely over time. If you think about what Jakob shared earlier of the five-second, five-minute, five-hour tasks. If you think about a five-year or five-century task.
[0:50:48] That would take something that would take humans. It's hard to even think about what that means. But probably you want a different kind of product experience. I also think you probably will want this to feel more like a sort of ambient, always present companion. Like right now, you can ask ChatGPT something. It can do something for you. But it'd be really nice to have a service that was sort of just observing your life and proactively helping you when you needed it and helping you come up with better ideas. And just I think we can probably push very hard in that direction.
[0:51:18] Neil asks, I love GPT 4.5. It's by far the best on the market for writing, and it's the main reason I pay for Pro. Could we get some clarity on its future, please? We think we're going to have models that are much better than 4.5 very soon, and for writing, much, much better. We plan to keep it around until we have a model that is a huge step forward in writing, but we don't think 4.5 is that good anymore. We'd like to offer something much, much better. Yeah, we are definitely not done with that direction of research.
[0:51:49] And yeah, we expect combining that with other things we're working on will get us dramatically better than 4.5 on all axes. Do you have any sense of timing to share about when you think we have a model that is dramatically better than 4.5 on this kind of task, like writing, and also anything about how far that's going to go? Next year, I think, is definitely what I expect.
[0:52:19] When is ChatGPT Atlas for Windows coming, asks Lars. I don't know an exact time frame, some number of months, I would guess. It's definitely something we want to do. More generally, this idea that we can build experiences like browsers and new devices that let you take AI with you, that get towards this sort of ambient, always helpful assistant, rather than something you just query in response. This will be a very important direction for us to push more on.
[0:52:49] Will you disclose the documents with the opinions of the 170 experts so that we'll have some transparency regarding the new justifications for model behavior? I will ask Fiji Simo how she'd like to handle that, but I don't know exactly what we'll be able to share, but I think we should do something there, and I think more transparency there is a good thing. Anonymous says, I've been a pro user since month two. As a researcher and fiction writer, I feel GPT helps me think clearer. I lost the question.
[0:53:20] Sorry, that was a really good question. Let me try to find it again. Clearer but not freer, has imagination become an optimization casualty? What do you think? I think it is definitely possible for our current systems. I think if you compare a model like 4.5 to a model like 03, I would expect that there will be trade-offs there. I think there are definitely like...
[0:53:54] Transitory as we figure out our way around these technologies. And so again, I expect this will get better. Yeah. I think there are going to be population scale. One of the sort of strange things I've noticed is people in real life talking to ChatGPTEs where they sort of use some of the quirks of things ChatGPT says. And I think there will be other things like this where there's this co-evolution of
[0:54:24] People and the technology in ways we can't totally predict. But my expectation is over time, people are much more capable, much more creative, think much more expansively and much more broadly than they do today. And we start, we certainly see examples of this where people are just like, I never would have been able to keep this in my head. I never would have been able to have this idea. And then we hear other examples where people say, you know, I've outsourced my thinking and I just do what this thing tells me. And obviously we're much more excited about the former than the latter.
[0:54:56] Can you help us understand why you build emotionally intelligent models and then criticize people for using it for accessibility reasons when it comes to processing emotions and mental health? Again, we think that's a good thing. We want that. We're happy about that. The same model that can do that can also be used to encourage delusions and mentally fragile users. And what we want is people who are using these models intentionally.
[0:55:20] The model's not deceiving the user about what it is and what it isn't. The model's being helpful. The model's helping the user accomplish their goals. We want more of that and less of anything that would feel like the model tricking a user, for lack of a more scientific word. We totally get the frustration here. Whenever you're trying to stop something that is causing harm, you stop some perfectly good use as well. But please understand, the place we're coming from here is trying to
[0:55:47] So I think already we are at a point where I think a lot of the
[0:56:39] The gap that stops present models from being able to perform a lot of intellectual jobs is more about integrations and interfaces than maybe raw intellectual capability. And so I think we definitely have to think about that, think about automation of a lot of jobs as something that...
[0:57:11] What will be happening over the next years? And I think this is a big thing for us to collectively think about. What are the jobs that will replace those? And what are the new pursuits that we'll all engage on? This is a question from me, not from...
[0:57:38] Not from the live stream. What do you think meaning will look like? What do you think the jobs of the future will look like? How do you think when AI automates a lot of the current things, how do you think we'll derive our fulfillment and spend our time? I expect, yeah, I think this is a quite philosophical question. I think it can go in many directions, but some things I expect, I think the high level goal setting, right?
[0:58:23] What pursuits we're chasing, that is something that will remain human. And I think that is something that a lot of people will derive meaning from. I think also just the ability to understand so much more about the world, the wide, the incredible variety of new knowledge and new also entertainment, but also
[0:58:54] Just intelligence that will be in the world, I think will provide quite a lot of meaning and fulfillment for people. Okay, rapid fire, two minutes. Shindy says, when GPT-6? I think, in some ways, maybe that's more of a question for you. In that, I think with GPT-5, we have...
[0:59:24] With previous models, like GPT-4, GPT-3, we've kept a very tight connection of how we're training new models, what are the products that we ship. And as I was just saying, I think right now there's a lot to do on the integration side. So for example, with GPT-5, it's the first time we really bring reasoning models as our flagship, our main flagship model.
[0:59:54] We're not coupling these releases and these products as tightly to our research program anymore. Yeah. I don't know either exactly when we'll call it that, but I think a clear message from us is, say, six months from now, probably sooner, we expect to have huge steps forward in model capability. Felix asks, is an IPO still planned and how will the structure then look like? Are there rules in place for increasing capital? We don't have specific plans or this is exactly when it's going to happen, but I think it's fair to say it is the most likely.
[1:00:24] You mentioned being comfortable with the $1.4 trillion investment. What level of revenues would you need to support this over time? What would be the largest revenue driver? It can't just be a per user subscription. Eventually, we need to get to hundreds of billions a year in revenue, and we're on a pretty steep curve towards that.
[1:00:50] I expect enterprise to be a huge revenue driver for us, but I think consumer really will be too. And it won't just be the subscription, but we'll have new products, devices, tons of other things there as well. And this says nothing about what it would really mean to have AI discovering science and all of the revenue possibilities that that would unlock. And as we see more of that, we will increase spend on infrastructure. OK, we are out of time.
[1:01:16] Thank you all very much for joining us and the questions. And we will try to learn from this format and iterate on it and keep doing these sorts of Q&A. Thank you very much.