Thibault Sottiaux

“When we asked questions related to why Codex runs in a certain way, Codex decided to debug its own systems. It connected to logs, SSH'd into research dev boxes, and analyzed ML instabilities.”

Thibault SottiauxHead of Codex · Leader, OpenAI

“Codex is really built for multitasking. There's this understanding that most tasks will just get done to completion. People on our team have figured out what Codex is and isn't capable of. There is a tricky thing in all of this, though: we have to relearn these capabilities with every model”

Thibault SottiauxHead of Codex · Leader, OpenAI

“We take a stance with the sandboxing that hurts us in terms of general adoption. However, we do not want to promote something that could be unsafe by default. As a dev, you can always go into your configuration and disable these settings if you want.”

Thibault SottiauxHead of Codex · Leader, OpenAI

“Yeah, I would say like, … skill that is trending … is typing fast. That is not that useful … Skills that are trending up is, you know, just great taste. Thinking about the user connecting to the audience that you're building for. A lot of ex-founders are incredibly successful right now, or like, you know, just like repeat founders. We're also finding this, like when we're hiring at OpenAI, like, you know, we have a lot of founders. I think we have more than 120 … founders currently at OpenAI. So it's like this mega startup.”

Thibault SottiauxHead of ChatGPT and Codex, OpenAI · Leader, OpenAI

“Yeah, yeah, yeah, merging some code every once in a while. Technically, like a lot of code is getting written for me to do all sorts of kinds of like analysis, like, you know, understand like trends, understand the business, understand the next feature, you know, how well our previous launches are doing. A lot of code is written by Codex for that. You know, I'm obviously not writing it by hand anymore.”

Thibault SottiauxHead of ChatGPT and Codex, OpenAI · Leader, OpenAI

“We are spending more and more compute on sort of like secondary monitoring. So you have, you know, you have all of the compute that is going into the primary system, you know, the agent that is like doing work. And then you have all of the compute that is going into monitoring, you know, the primary agent to make sure that it's like not taking too high risk of actions … If anything looks like, you know, maybe it's getting prompt injected and just like, you know, intervening with that. And the majority of our investment on the API stack is like, you know, actually going into the safety stack.”

Thibault SottiauxHead of ChatGPT and Codex, OpenAI · Leader, OpenAI

“I don't know if that's surprising, but it is, you know, as I said, a lot of ex-founders doing a lot of bottoms up, you know, very exciting ideas. You know, for example, decisions API, you know, came together very quickly. We realized like we have a really good model with Luna. We can just do some, you know, constraint sampling and then, you know, have it, you know, shipped as like a different shape to the Responses API. And then, you know, when we do that, you know, it's like, it's faster and it's quite delightful for some cases. And it's just, if you were to like, … the hood, you realize like, okay, there's this … channel that came together. You know, initially, it was like four people kind of hacking on it on the weekend.”

Thibault SottiauxHead of ChatGPT and Codex, OpenAI · Leader, OpenAI

“…me five minutes before the live demo, which was actually … stressful because it was like, … production is down. And it's like, oh, well, you know, it's like, … want me to try and fix it? And I'm like, I don't think you're there yet, you know, a little dot, but, you know, thank you for trying. And then, you know, then, you know, I got in touch with like the engineering teams and we started to look, you know, what was going on. And, you know, they, it's like fixed by now. But having this thing that just understands like, okay, there's a pretty important thing happening. It's called Dev Day. There's like this production system, there's a live demo. It's probably using this production system. So these two things are connected and this is happening in five minutes. So I should ping him because like, he probably wants to know about it. I think that's quite remarkable.”

Thibault SottiauxHead of ChatGPT and Codex, OpenAI · Leader, OpenAI

“I think like having to set up and fiddle with your loops and, you know, figuring that out is something that, you know, maybe people got excited about, but I don't think this is the way that it's going to work. I think the way that it's going to work is how we're positioning and what we're shipping with dots, where you have an incredibly smart agent that works 24/7, understands your goals, understands your preferences, learns from feedback. And it's like, … not perfect. I will learn very much from making this available to our pro users. But over time, you just want … system that learns from what you want to achieve.”

Thibault SottiauxHead of ChatGPT and Codex, OpenAI · Leader, OpenAI

“But, you know, seriously, like … research, long persistent, long horizon tasks, this is something that we've worked on, you know, for more than two years. … And it's like, you know, it's just memory works very, very well in … and so we've put a lot of research in that. And, you know, all of this coming together on top of like the Codex harness and also the ability to just like run 24/7 with, you know, good productivity. That's what we launched with dots. And so it's the work of like many, many months.”

Thibault SottiauxHead of ChatGPT and Codex, OpenAI · Leader, OpenAI

“It has access to some production systems with guardrails, you know? So we built, we talked about specialist dots, but specialist dots is just like, operate with additional guardrails, additional monitoring, and then, you know, on their own hardware. So we actually run some on … The fundamental difference in the way that we've built dots is like, the harness does not run on the machine, you know? And then this is like, maybe people have not realized. It's like, it can connect to as many devices as you want. So it has its own computer. You can actually connect it to your own laptop as well. Over time, you might connect it to like, 10 different devices and it can control it all, a little bit like an octopus.”

Thibault SottiauxHead of ChatGPT and Codex, OpenAI · Leader, OpenAI