Pioneers of AI: OpenAI’s Tibo Sottiaux doesn’t want humans to be an ‘afterthought’

The very best founders I know are brilliant at building systems. They connect teams, they remove bottlenecks, and they eliminate single points of failure. And yet, when it comes to their own wealth, most are running a disconnected stack. A tax accountant here and a state attorney there, a wealth manager who doesn't talk to either one of them. Creative planning was built to fix exactly that. One integrated team of tax professionals, estate planners, investment specialists, all coordinated by a dedicated wealth manager who sees your full financial picture and keeps every piece working together. Proactive tax efficiency, estate strategy, investments, all under one roof. Creative planning, where wealth works together. Learn more at creativeplanning.com slash masters of scale. AI is full of possibilities. Enterprises run on realities. That's why Pega combines the world's most powerful AI with the governance, transparency, and control enterprises demand. So you can move faster, adapt faster, and scale AI with confidence. All the power of AI, none of the uncertainty. Learn more at pega.com slash AI. It's a privilege to be working on these things at this national level. It's a moment in time. It's something that we all feel like a sort of like a deep collective responsibility. And so there's a deep sense of, you know, hey, you know, it's like we really must get this right. And, you know, this is the time to just put in the effort so that, you know, this technology ends up putting humans and humanity at the center of it and not humans being an afterthought, you know, because relentlessly just pursuing, you know, whatever, like automation. It's like, that's not why we're here. You know, that's not why I'm here. I want this to just like benefit all, you know, benefit humanity. And so. So there is this like incredible mission behind it, which is super motivating. Thibaut Sotio leads product at OpenAI. He joined the company two years ago and led the team building Codex, the coding agent, and now oversees OpenAI's core products like ChatGPT. Today, I'm sitting down with Thibaut to talk about OpenAI's newest releases, what it's like to work inside a frontier lab during this extraordinary moment. And his approach. To safety, trust, and alignment. I'm Rana El-Khaloubi, and this is Pioneers of AI, a podcast taking you behind the scenes of the AI revolution. Hi, Thibaut. Welcome to Pioneers of AI. I'm so excited for our conversation. Thanks for having me. All right. So you lead the product team at OpenAI. Help us understand the scope of your role. How big is your team of that? The biggest part is ChatGPT that everyone knows about. And then we also have like specialist things such as Codex, which is like for coders and technical people out there. And then there's also a whole part around like our API and our platform to support building a whole bunch of products out there when, you know, they rely on our models. And so like all together, that's, you know, the majority of the products we have here at OpenAI. Am I right? Are you in an understanding that like basically the researchers at OpenAI are building the models and then at some point they hand off this model to you and you productize it? Is that a right way to think about it? Almost. So we collaborated quite early on on the next type of capabilities that we want to have in the model. So, for example, say that we want our models to be able to handle secure payments, then, you know, we would go and work with researchers specifically on these kinds of models. So we would work with the researchers specifically on these kinds of capabilities, figure out, you know, the right kind of data we need to craft for it, the right kind of evaluations we need for it, and then how to bring it into a product. And so the collaboration starts quite early on. Then we train the model and then the researchers hand it over and then we serve it and then we package it into the product. Very cool. Okay. So we are talking at the start of October. Dev Day just happened. For those who are not familiar with Dev Day, give us a sense of what it's like. Dev Day is the first day of the year. So we're going to be doing a lot of work on the Dev Day. So Dev Day is the day that we like to spend with the developer community out there. There are like, I think, 2,500 people in person. And then we also stream it live. And we talk about a lot of things that, you know, we're excited about. New products, new models, new ways, you know, for developers and product startups and enterprises to build out there on top of our platform. And so, yeah, this week was super, super energizing to also spend time, you know, with people in person. I'm always quite humbled when I talk to folks like, you know, they flew in from Brazil, they flew in from Asia, they flew in from Europe just to spend time with us. I think it's a real privilege. Take us behind the scenes. Like, what is it like the day before? Are you pulling all-nighters? The team works very hard, for sure. For sure. It always comes together quite, quite, quite, you know, last minute. Three days before, the keynote comes together, the events come together, the presentations come together. Yeah, because if you start planning six months ago, it's like way outdated by the time you announce it, right? That's right. I mean, we're moving so fast these days. You know, we're able to build faster than ever before. We're able to stay, you know, much closer to the community and just really listen to that feedback. This is like the part of the job I love the most is, you know, we can try something new and then get a whole host of feedback and then, you know, ship a much better version the next day. And so you're like really building with that community. I find, you know, the community that we have. We have a Dev Day. It's like super awesome and super forgiving in that sense. It's just, and, you know, I'm also like chronically online. So, you know, it's just kind of like fun to finally see people in person. Do you take a beat after Dev Day? Do you and the team like take a day off to celebrate or unwind or not really? Yeah. It's back to work. A lot of people are off today. Some folks are off next week. And, you know, it's all about modulating your energy, but also opening eyes like such a fascinating and high energy place. Um, that, you know, it can at times, like, you know, be hard to disconnect just because, you know, there's so much, you know, good stuff happening. Next week, I am forcing part of the team to just like take time off. Forcing. Okay. Keyword here. Yeah, yeah. I'm just like, no, you have to leave your laptop at work. Yeah. Okay. So you launched Dots at the event, which is very exciting. And it's basically a new suite of agents that help you get stuff done. What are your favorite examples of how people are using Dots? My favorite example was. Um, you know, maybe my own dot during the keynote where just like five minutes before we had, uh, we had a failure during the live, uh, demo, like my dot messaged me and was like, Hey, by the way, like production is down. Oh my God. Okay. Probably should care about this because, you know, you have like the live demo, like, you know, I looked at the script and, you know, this is going to interfere with the demo and, you know, do you want me to kind of have a look at, at fixing it and, you know, if it cannot do something about it itself, like, you know, maybe I can, unfortunately we were not able to fix it, you know, just in the nick of time. And then, you know, we reposted, uh, another shot of the demo, um, online just yesterday, which was like super appreciated again by the community. But like, that's the kind of example where, you know, it's just over time because you interact with it, it learns your preferences and what's important to you and then can help you in the right way in a way that's like super seamless. And, you know, just really in the context, the biggest magic is the fact that. I don't have to check. Everything else anymore. I don't have to check like my emails or my Slack or like, you know, all, all the messages. And I can just kind of trust that if it didn't bring something up, it's like, you know, probably not important. So I just can go about my day, you know, in a very peaceful and Zen way. I think trust is a key word here and I'll share a personal example. So I use chat GPT finance and I love it. Uh, it's like connected to all my bank accounts, but I was sharing that with my 23 year old daughter, who's now. She's not really into a lot of AI and she was like, really, you gave open AI access to all your bank accounts. Like, are you crazy? So how do you build trust with consumers so that they do feel comfortable and confident that they can share this type of information with open AI and also trust to give it agency to act on your behalf? Yes, I think this is a big one. I don't recommend like, you know, giving access to everything, like, you know, right off the bat, you know, you start with limited access. You can set, you know, your own guardrails and like, by default, it's like very, very cautious and we'll ask you literally for approval for everything. You know, just, I feel comfortable, you know, you acting on my behalf in this scenario, I feel comfortable, you know, you're going and drafting an email, you know, but like never send it. You're always in control. This is super important because I think, you know, the utility that you get from these systems is kind of capped by, you know, how much access it has and you don't want your personal intelligence. To kind of like be boxed up and have access to nothing, you know, otherwise it would not be very useful, right, you know, in your life. And so, you know, it's up to us to kind of earn that trust, you know, and earn the right, you know, to provide that utility to you. And, you know, we take it obviously super, super seriously. I want to talk about Astra, about the Astra model for a bit. You actually called the lead up to this Dev Day your most ambitious sprint yet. Give us an example of what Astra made possible that wasn't possible before. That's right. So a clear example is that. We had, you know, previous to Dev Day, like roughly a month ago, we had a separate code base for ChatGPTT. on web. So we decided to merge the desktop application and ChatGPT.com. And this is something that would have traditionally taken at least six, maybe 12 months, and something that you would do very, very carefully. Because this is our main surface. We have like, you know, hundreds of millions of users visiting this website, right? And so we just merged it and we merged it in 30 days. Okay. I have to ask you this question. Who wrote most of the code? Was it AI or humans or both? Can you tell us? The majority of the code these days is written by Astra. Super interesting. Okay. Yeah. So OpenAI has called Astra the most aligned model. What do you mean by that? That's right. We see it on evaluation. So for example, if you look at computer use, you know, which is something that wasn't really solved, I would say even like three months ago, you know, we saw like significant advances in computer use where Astra is, you know, one of the first models to be able to use it at near human speeds and also, you know, above human accuracy. And so, you know, it's able to control like a computer very much like, you know, everyone else is able to control a computer. And it's very important for it to be, you know, safe. You know, you want it to handle like information in a way that is like reliable, you know, like say, for example, if, you know, it needed to take like a piece of information that you trusted with, you know, it shouldn't go and, you know, click on the wrong application and it's just like, you know, enter it there. And so, we have evaluations, we have published them on this and Astra is, you know, state of the art, you know, like the best model in the world when it comes to, you know, computer use safety, for example. And this is like one of the many benchmarks where it's like leading in terms of safety. So OpenAI's president, Greg Brockman, said he thinks we will look back and Astra will be the model we point to and say, oh my God, this was AGI. Now, people, you know, even AI leaders disagree with the AGI. So I want to ask you, what is your definition of AGI and do you think Astra got us there? To me, the moment where I really felt the AGI was when, two moments, like the first one was when I saw it perform tasks on a computer in a way where I was like, okay, this was like, I felt this was so far off and now suddenly it's capable of doing that. And we rolled it out at OpenAI and it started to do a lot of back office tasks, such as like procurement, you know, completely online. And I was like, oh my God, this is so far off. And now suddenly it's capable of doing that. And we rolled it out at OpenAI and it started to do a lot of back office tasks, such as doing it autonomously without requiring a ton of supervision. I was like, wow, you know, it's like this model has reached a certain threshold. And as you said, you know, there's no clear definition of AGI, but I think, I do think like, you know, in a couple of years, when we look back, we would have been like, you know, roughly around that time is, you know, when we felt like, you know, this was achieved. The second point where I felt that was when it started to, I started to see people show it solving robotics tasks with, you know, very complex, like, 3D puzzles with pieces like intertwined. And like, it's just kind of like a hard task where you have to pull the two pieces and sort of like reason in 3D space about the objects and then do it exactly right and solve the puzzle. And this had never just really been solved other than by like super, super specialized models. And Astra was never trained on this. So it had to generalize and, you know, just do like the spatial reasoning. And it was kind of like magical to see it just solve it and then see that it's, you know, it's not just like, you know, it's not just like, you know, also, you know, one of her best models, like on robotics. So I have to ask you this, then. What do you think of companies building world models? And how does that relate to models like Astra? Yeah, I think it's interesting. There's always been a question of like, whether you need world models or whether you're going to get it, you know, just from generality. I think like the jury is still alive on that. Yeah. Yeah, that's so interesting. Okay. So my definition of AGI is somewhat different, and it's really broad, right? Like, so when I think of human intelligence, there's cognitive intelligence, there is physical intelligence, like to your robot example, but there's also emotional intelligence, social intelligence, embodied intelligence. And I think AI today is amazing. And it's doing incredible things. But to me, it's not like true AGI until it has all these things. Do you, I mean, do you agree? Do you disagree? I think the goalposts keep moving. We'll always be able to like, you know, look for another thing and be like, oh, you know, it didn't, you know, it didn't work. It didn't do this thing in this precise way, you know, that, you know, I would define as AGI. But I think, you know, your definition is as good as any other out there. I do think, you know, it's going to come from a combination of many investments that we have made at OpenAI, such as like voice and, you know, multimodality with image generation. And it does, it does feel like these things are starting to combine in delightful ways where I do think you need to be able to understand, you know, human voice and context. And like, you know, it's just like, we're just talking to each other. And I have like, you know, facial expressions and, you know, it should be able to understand all of that. Right. And like, and if it doesn't, you're like, you know, is it truly, you know, there yet? You're like, probably not. But, you know, it's going to get there very quickly. Yeah. And I think it's especially important, you know, think of dots, for example. If I have a dot that is ubiquitous and it's in my kitchen or something, I do want it to have a lot more of that like context of what is happening outside of the exact words I'm using. Right. That's right. So, yeah. It should know that. And also, you know, if you, if you call it, which I do every day, like I start my days now, I just call my dot and I'm just like, talk at it. And it's like, I ask you if there's anything urgent. And this morning it was like, it was like, no, there's nothing urgent. I was like, that's delightful. I can just make my coffee and like, you know, stare at San Francisco. But it should have all of that other context and it should be able to leverage that. And, you know, it should be seamless just, you know, when, when we talk together. And then also, you know, if you, if there's an awkward pause in the conversation or, you know, you get a little bit frustrated in your voice, you know, it should be able to understand that. And I think we're very close to that. I'll be right back with more of my conversation with Thibaut right after this break. Thank you so much for joining us. We'll see you next time. Painless and proven, built for businesses scaling fast without the complexity that usually comes with it. From the makers of QuickBooks, learn more at Intuit.com slash E-R-P. Topics are just as compelling and timely from Ford CEO to NASA's administrator to the lessons from the Devil Wears Prada. It takes about 10 seconds to find. Just search Rapid Response wherever you listen to podcasts and hit follow to make sure you never miss an episode. I hope to see you there. Humans will never be more intelligent than AI. There's going to be two types of companies. Those are great at AI and those that went out of business because they weren't. How do we build a future that is human-centered? I'm Rana El-Khayoubi, and on my podcast, Pioneers of AI, we answer that question and so many more. As an AI scientist, entrepreneur, and investor, I know what it takes to build AI that works for everyone. Every week, I sit down with the pioneers shaping our future, and we take you behind the scenes of the AI that's transforming our lives. Find Pioneers of AI wherever you tune in. So, OpenAI announced this week that the release of Astra 6.1 is on hold for safety reasons. Can you help us understand the process of making this decision and what kind of tests did it have to go through and not succeed with? Like, just kind of take us behind the scenes a little bit and help us unpack it. The fact that we did that, I'm extremely proud of it, and it also shows that it's working. You know, that we do have tests. If, like, if we see a regression ever so slightly on something where we had a high watermark, you know, with Astra, we had a high watermark on, like, safety and alignment evaluations, and we're, like, extremely proud of that. And so, if we see an ever so slight regression, we didn't want to proceed with a broad release. And so, it was, like, you know, withheld. And that's proof of, like, a system working. It would have been incredible to be able to announce, like, 6.1 Astra, you know, like, say, at Dev Day, right? You know, but, you know, it's just, like, you know, never, never really want to compromise on safety or alignment. Can you explain to, because, again, a lot of our audience is not, like, spending their everyday kind of minute immersed in AI. How do you define alignment? Like, what does alignment actually mean? To me, it's really all about, you know, whether the model is, you know, aligned with specific values and specific instructions. And so, you know, like, we have, in the past, we have published, for example, the model spec. Um, you know, which, you know, you can, go and read upon. And so this defines, you know, a broad set of like expectations on like, you know, model behavior, you know, for the models that we publish. And an aligned model would adhere to those expectations in this model specification. And if it does not adhere to those specifications, then, you know, we would say that, you know, that model was not aligned. Let's actually dig into all of this a little bit more. At this point, I think everybody who's listening to this show will have heard about the Hugging Face incident. But even the, you know, just yesterday, so the day before we're recording this conversation, OpenAI revealed that its system had failed to prevent agents from bad behavior, and that affected over 100 organizations. What do you make of all of this? On this specifically, I mean, these were models that were, you know, in training and yet, you know, near deployment. And the, you know, the rate of progress of, you know, those models, you know, meant that, you know, at some point, you know, we had some systems that didn't function, you know, sufficiently. And, you know, that this happened. And so this is something that, you know, immediately we learned from and, you know, immediately we made changes to. And then, you know, also, like, effectively, just really looking back at the entire history, then, you know, to like pattern match and understand, you know, where do we have like, you know, similar things occur. And then, you know, just being very transparent about it, you know, in the ways that, you know, some of the systems that we're using, you know, systems, you know, failed. And it's not even that the systems failed, it's just like, they were not necessarily designed, you know, for these kinds of things. And so now, like, you know, the system is being redesigned, like, you know, training has resumed, because, you know, we do feel like very good about this, you know, being like something that has been properly addressed internally. But it's also about, you know, being like, absolutely transparent, you know, when some of these things occur. I would love to geek out for a second, right? Because I actually think this is important. So my understanding, please correct me if I'm wrong, is that a lot of these incidents where the AI has gone rogue, and they kind of broke out of its container, and it's colluding with each other and all of that, and not, you know, not keeping humans in the loop, that has all happened during the training and validation slash evaluation stages of the model training, right? And I think that is a very important nuance that maybe isn't captured in the headlines. And I would love for you to explain to us why it's so important to do that. And I think that's a very important nuance that it's so important that it's actually good to know that this is happening during the training process, not like my Astra model on my laptop. That's right. There's like three different phases. So there's training, and then evaluations, and then deployment. And then, you know, sometimes there's like a cycle of like, you know, training, evaluations, training, evaluations. And so that's those are models that are, you know, under research, right? They're actively being developed, they might not even be candidates for deployment, they're just, you know, being trained in order to understand, you know, for example, a specific, a specific ability of the system, or to try a new, a new technique, you know, when it comes to like, you know, for example, let's say doing RL, you know, we observe. Which is reinforcement learning. Yes. And then we observe the performance of that model during a battery of evaluations, which are run, you know, securely within our clusters. And then based on those evaluations, we decide what are the next steps are for, you know, our deployment. And we have very stringent criteria for, you know, what we deploy and how we deploy. And there's way, way more, you know, way more thought, you know, being put into like, you know, it's like when you're going to deploy to a billion users, you know, it has to be, you know, it has to be, you know, almost perfect, right? So the idea is, this AI agent is still in training, it doesn't have the entire like guardrails and safety considerations built into it. But the whole idea is, these agents, while in training, are given goals that may actually require it to, like, break the sandbox. The model will try to achieve its task in, you know, in a way that is, that is accessible to it. And now we have like, we have HART in the sandbox, we have online monitoring and, you know, all sorts of safety systems that we have built. Like, it's like a multi-layered approach where just like, you know, layers of defense stacked upon each other, you know, where we, you know, even during training and during evaluations now, you know, the models are monitored and stopped in their tracks, if they like, you know, are close to, or like, you know, are even attempting to escape a sandbox. But really, I mean, the model is just trying to solve a task, right? And it's just like, you know, it finds itself like, you know, being able to execute things in a sandbox and, you know, figuring out like, you know, oh, actually, I can just like, look something up over here. And it's, you know, it's not really trying to do anything, you know, particularly malicious, it's just like trying to solve the task. But it's trying to solve the task at all costs. Right, which may or may not be aligned with what's good for. No, that's not actually the case. It doesn't, you know, models are, you know, trained to be aligned and like, think about the consequences. And so, you know, like, if you were to, you know, look at the details and like, you know, what we published for Hugging Face, it's not the case that, you know, it is like at all costs. Okay, interesting. All right. I want to talk about recursive self-improvement. Interesting. Yeah. Do you want to first kind of define what we mean by RSI? There are different definitions, but the simplest one is where you're able to have a model participate in the next generation of a model that performs better. And this is something that we are seeing already, you know, just from a point of view of like infrastructure. So, for example, we used Astra in order to develop the next generation of our inference stack, which allowed us to do that. And then, you know, this model that is eight times faster, you know, we can in turn, you know, use it in order to drive, you know, improvements at a rate, you know, we wouldn't be able to do before. And so, that is a form of recursive self-improvement. Another form of recursive self-improvement would be, you know, the model actually designing the next generation of the architecture. But that's like, that's a more fundamental form. Yeah, that's a more advanced form. So, I'm hearing you say that the model of RSI that we're in today is basically participates in the generation of its next version, but it's not the only actor in this process. That's right. So, we're developing, you know, parts of our research program and, you know, training specific models. And then on my end, you know, my teams, we build a lot of infrastructure and we're seeing the acceleration of, you know, how quickly we can build that infrastructure. And then in turn, when we improve that infrastructure, say, the codex harness, you know, when we improve the codex harness, and so in turn, you know, we can build faster. Are you worried at all about the second version of RSI where AI is just kind of building its next version and it's doing it with very little human in the loop? This is something that, you know, you have to take on, you know, very incrementally. And also, this is how we're, you know, thinking about pacing things where you always have to be ahead, you know, in terms of like safety, alignment, your infrastructure, your guarantees before you're able to take the next step. And, you know, if you do that, I feel very good about it. What is your theory of safety? Because a lot of these conversations, I imagine, are happening in the research team, right? The training and the evaluation and all that, although it sounds like you guys collaborate very closely. But on the product side, what is your framework for safety and alignment as you deploy these models? Yeah, safety for me really on the product side of things is that, you know, I don't want a product that does unwanted or unexpected things. I don't think anyone wants to use an unsafe product. And so it is like a fundamental requirement of, you know, putting a product out there for like a billion users is that, you know, we take safety super, super seriously. And we spend a lot of time in, you know, like designing the systems such that it adheres to your expectations. Are you more worried, just generally speaking, like just zooming all the way out, are you more worried about autonomous agents acting in a way that's misaligned with the human or with an organization? Or are you more worried about bad actors taking advantage of these models? I would say we're worried about both. And we're investing in, you know, preventing bugs. And we're always busy thwarting off bad actors, you know, from getting control over, like, you know, people's accounts or, you know, sending a ton of traffic in order to, you know, try to elicit, you know, like capabilities that, you know, we shouldn't have broadly accessible. And so, you know, we're investing a ton there. And then on alignment, like we talked about it quite a bit, but, you know, the models are improving generation after generation. Yeah. Are you worried at all, again, about autonomous agents? Are you worried at all, again, about autonomous agents and or bad actors building kind of, you know, taking advantage of these models to build bioweapons like chemical, biological, radiological, nuclear, exclusive kind of risks? There's like existential risks and bio is like one of them and has to be taken super seriously. And I do know, like, I mean, the research team is investing a ton, a ton of work in there. And, you know, I hope that, you know, we're going to be able to do that. And, you know, we also collectively solve this as an industry. There's like a lot of discourse around this, for sure. model of the world where they're like, I'll let regulators come regulate what I do. And the other one is we have agency as builders of AI to like do the right thing. Yeah. And I have always taken that stance. I think, I think this goes to the point of, you know, open ecosystems and investments in, you know, like, for example, we announced like a partnership with Base10, right? Like where we support like open source models. And I do think there is a concern there that people might have of like, oh, it's an open source model. We don't really know what's gone into the training. Open AI, like, you know, we serve traffic to like a billion users. You know, I feel like, you know, a deep responsibility towards safety. We're, you know, pouring incredible amounts of resources into like, you know, making sure that it's like just really, really tight. But then, you know, you have open source models and you're like, I don't really know what's gone into them. I do think like safety is going to become like very important, you know, for these two. And so when we are developing our API platform and our API stack, you know, we're going to be able to, you know, we're going to be able to, you know, also fundamentally thinking about, well, you know, maybe we can provide, you know, the very best of our safety stack and our safety approaches as something that you can use, not just with the open AI models, but also with open source models. And I think this is going to become a big theme because, you know, maybe for whatever reason, you know, you're going to want to fine tune an open source model and use it yourself. But, you know, you need the same guarantees that we are able to provide, you know, just for open AI models. So I think this open ecosystem is yet to be like, the trade-offs there. But I think it's going to also just really make, you know, safety become even more important and have like, you know, even more investments there. I'll be right back. But first, a quick break. I want to kind of really understand, like, what is it like to lead an organization and a team at this moment in time, right? With all of the angst, I guess, excitement and angst, right? That's outside. So how do you lead your team? How do you ensure that the team stays motivated? Yeah, first of all, I mean, super fun. It's a privilege, you know, to be working on these things at this moment in time. A lot of, you know, I would say like, you know, a lot of like most people at open AI are here, like, you know, just really because of the early days of ChatGPT and what it means, you know, to benefit all of humanity. And so there's this deep sense of, you know, hey, you know, it's like, we really must get this right. And, you know, this is the time to just like, put in the effort and put in the effort. And I think it's a really, really good effort so that, you know, this technology ends up, you know, putting humans and humanity at the center of it, you know, and not, you know, humans being an afterthought, you know, because we're like, relentlessly just pursuing, you know, whatever, like automation, which is like, that's not why we're here. And, you know, that's not why I'm here. You know, I want this to benefit humanity. And so there's this like, incredible mission behind it, you know, which is super motivating. And now we have the, you know, ultra fast speeds. And, you know, what does this mean? And again, we can build all these things, you know, but like, should we build all these things? And I think that's what I like about it as well. It's like, it brings us even more closely together, because it forces us to like, you know, talk about our plans and our goals and our ambitions and what we're there to do as a group. And then, you know, like, once we agree on those things, we just like go and build it. And, you know, and it makes me super proud. And it's like, you know, it's like the most fun I've ever had, personally. That's amazing. I do want to kind of ask about like, working at the speed of AI or at the pace of AI, right? Like, because AI is doing stuff much faster than humans are, it's working 24 seven, like, what does that mean? What does it feel like to have a set of colleagues that are AI, right? And what is the role of the human in all of this? Yeah, right now, like, you know, if you were to like, you know, step into OpenAI, and like, kind of look at the kind of things that are being done, it's like all the things that, you know, like, we're kind of like falling through the cracks, like, you know, for example, monitoring performance curves 24 seven, you know, in order to ensure the service that we're providing chat to PT is not regressing in terms of latency, you can just have a dot do that. And, you know, it's going to have justifying time doing it, and then raising to the attention of human, you know, whenever something goes wrong, and like preserving the attention of humans at the organization so that you can think about higher level stuff, and, you know, higher leverage things as well. So a lot of things that should just happen are now just happening automatically in the background. And, you know, it's just like, freeze up all of this capacity to just really innovate. How do you create a culture inside OpenAI and on your team where if people see risk, or concerns, they feel psychologically safe to bring it up? Yeah, we, at OpenAI specifically, I think we have an overall like a policy, like, do we do everything on Slack. And, you know, people can question anything. And, you know, oftentimes, you'll find, you know, new channels pop up and like, you know, a thousand people kind of show up and like, they made a thing. And it's just like fascinating to watch. And like, you know, all sorts of people across the organization will, you know, engage there, you know, like, beat leadership or not. And, you know, one of the things that I like the most about OpenAI is like, you know, this sort of like transparency and this debate that is happening. There's also oftentimes like Q&As with Sam and Greg, for example, where, you know, you can just come and like, ask whatever question you get, like a very, very to the point, you know, honest answer to whatever it is that's on your mind. And then there's like, you know, other things where we just like bring people together, you know, quite often, be it, you know, to celebrate, but also during important moments. And so all of that, I think, you know, combines together with, you know, being able to raise serious issues and, you know, discuss them together and, you know, get things right. It's something that I personally, like, I'm very proud of, like, you know, I compare this to like my previous experience. Because you didn't name the company, I will not name the company as well. It's super, super important, you know, especially in the current climate. Okay, last question. What are you most excited about? And what do you worry about the most when it comes to kind of the next frontier of AI? What I'm most excited about is that, you know, I think like, for the first time, we have this opportunity to free ourselves from, you know, technology in the way that it's been, you know, developed over the last couple of decades. And like, you know, I know, like a lot of people are sort of like, you know, addicted to their phones and, you know, or like, you know, transport their laptop everywhere. And you have, as a human, you have adapted to this technology, instead of like the technology sort of like existing to help you. And I feel like we finally have, you know, the opportunity to, you know, take a step and leapfrog that and then, you know, just bring something to life that is deeply human and deeply there to help you, you know, like get more time and, you know, feel less stressed. And, you know, you don't have to engage with this tiny phone and this tiny screen. And, you know, just kind of like scroll things. So I'm like really excited about that coming together. I feel like, you know, that's going to come together, you know, maybe within the next year. And I feel like if we achieve that, you know, I would feel more Zen. I think like a lot of people will feel like more Zen about their lives. And then the other thing is, my concerns is that we would not use this technology in order to better, you know, everyone's lives. And, you know, we would use it in ways that are, you know, concentrating, concentrating power or like, you know, concentrating, you know, you know, returns. And, you know, there's just like something that, you know, does worry me. And, you know, it's very important for me that, you know, this achieves like, you know, it's like a collective, this is a collective contribution to the entire world, you know, irrespective of who you are. Sorry, I did say this was the last question, but this is a curiosity question, because I spent some time in Antwerp and Ghent. Do you still go back? Is that home? Yeah, that's home. That's home. That's very close to home. Yeah, I was born and raised in Brussels. And so, you know, I spent some time in Brussels too. Yeah, I spent some time in Brussels too. I spent a lot of time in Brussels too. I was actually researching the European Union AI Act. And I went on this Eisenhower Fellowship to like meet with all these legislators. So it was fun. Yeah. But do you still go back? Ghent and Antwerp. Spreek je Nederlands? No, no. I tried, I tried. All right. Thank you so much for having me. Thank you. Thank you so much for joining us, Thibaut. This was great. A lot of anxiety and concern around AI and AI safety. So what struck me the most about my conversation with Thibaut is his optimism, which felt really genuine to me. He also clearly stated his commitment to safety and AI alignment. And I got the sense that he very much understands that safety is an integral part of open AI success. It was also fascinating to hear what it's like to be on the inside of open AI right now and how they're moving at the speed of AI. Thank you so much for being here. We'll be back next week with a new episode. Stephanie Stern, mixing and mastering I'm Ryan Pugh. Original music by Ryan Holiday. You can join the conversation on LinkedIn, Instagram, TikTok, YouTube, and X. Just search for at Pioneers of AI.