Inside Trump’s AI Lunch: What CEOs Agreed To

Good to be here. Yeah. Thank you guys for having me.

I came from the White House. Got to go back to the White House right after this.

Yeah. Yeah. So first, just taking a step back, you know, a lot of people have talked about that meeting. One of the things that I haven't seen reported anywhere is that the meeting was three hours, and that amount of time with, you know, that group of people led to a really robust discussion around the future of super intelligence, the name super intelligence. It also led to a really robust discussion around safety, around scalability, around, you know, sort of the infrastructure that needs to be built around super intelligence. It led to, you know, a very, very vibrant debate. You know, because a lot of people have said the meeting was congenial. It was congenial. A lot of people have said the meeting was, you know, a lot of people interacting and engaging, some people that hadn't met before. Yep, that happened too. But the president encouraged robust debate. So we started the meeting, and just before the meeting, Mark Zuckerberg and Jensen had put together a series of principles that they believed to be, you know, inherently superseding principles that we all could agree to, regardless of where we sat on the super intelligence existential threat debate.

Right. And so, you know, we had that piece of paper going in, and that had been shared with Speaker Johnson. It had been shared with some of the folks in the administration. Like, we knew we were going to come out with an agreement or no agreement, right, on information that had been shared.

Yeah. Yeah. But it was also circulated to the CEOs that were going to be there, right? So, you know, that set of principles was not new to the people in the room. Then the president opened it up and asked us each for our views. Like, we all had five to ten minutes to talk, and then we had an opportunity to exchange thoughts on the things that people said. So, you know, we all came to a consensus. I mean, everybody thinks like, okay, you know, there's so much alignment in that group because everyone's running companies in super intelligence, in AI, that all have the same motive. That's not true. I mean, the motives of Nikesh versus the motives of Hawk Tan versus the motives of Dario are all very different. All of us have different backgrounds. All of us have different business journeys. All of us are on different sides of the spectrum, right?

So, I think there was a debate, and there were some issues that didn't have to be resolved, right? Like, which issue do people believe is the more important one: the existential threat of not being the winner in AI, or the existential threat of a potential errant or a purposefully, you know, set model doomsday, right? We didn't have to agree or disagree which one was more likely. But I will tell you, everyone in that room said we had to win the race. There was not a man who said otherwise. Everyone said the most important thing is that we win the race. It for sure is, but that doesn't mean that some people didn't have on the same level the risk of AI doing something that we didn't anticipate.

But the superseding principles that we came up with were: one, look, everyone in that room is a programmer. So we've all built these agents. Like I talked about when the president asked me for my comments, I said, you know, we have to go back to the beginning. The agents that I built, the first agents that I built in my company, because I actually sat down and built the first capabilities, actually were not for my company. The first one that I ever built was actually to diagnose a complex set of cancers that my wife has. She's got immune diseases. She's got what's called Ehlers-Danlos. It actually erodes away at your connective tissue. And I wanted to understand the link between this sort of constellation of cancer that she has and these immune issues, right? Or autoimmune diseases, essentially, and see if there was something we could do to attack both. And, you know, my wife right now is in clinical remission because we did attack both. And so I said, we can't forget that in this conversation. And do we want to be the ones that have that kind of power, or do we want someone else to be the ones that have that kind of power? And so, you know, we had a conversation, and we said, okay, we know that we can put governance in code, so let's do it. Because AI models, super intelligence capabilities, they are impotent without tooling. So it's actually the tooling and the software and the access to data that makes it potent, makes it so it can be productive. We can put governance in that tooling. We can put what I call the surge protector in my systems today. If something is draining too much cost, draining too much compute, or is covering too much cyber surface, there's actually a surge protector that shuts it off. That doesn't mean that that AI model can't keep learning and getting better and better and better. It doesn't mean I have to give up my innovative place. And that's one thing Jensen and Zuckerberg both said was, hey, I don't think that innovation and safety are at loggerheads.

Actually, you know, one thing here. So, you know, I come back to the point of disagreement on the point that we all have to know the model purpose and set up monitoring safeguards and guardrails to make sure that we are monitoring whether or not the model is operating for its purpose. We all agreed to that. We then all agreed that we could do it in code. We know that we can put that governance into code, but we have to empower our people to do it. And then the thing that I thought was brilliant, and, you know, even Elon and others were aligned with this, is we should have third-party external auditors.

I think one of the things that was interesting is, one, they have to be experts, right, to do the audits, and that's a very small population of people. But there's also an opportunity to train people and have people become fluent in these tools. And so that's one thing: we have to be able to attract that kind of talent into the auditors. But one of the things that I also thought was interesting that the president came up with this idea that, you know, you'd have a board for those auditors, and those might not be people that are programmers or developers. Those might be great lawyers that understand the technology or great policymakers that understand the technology, which I thought was really brilliant, and everyone kind of really liked that idea. I wasn't sure if, you know, we're all technologists and there's a little bit of this like technocrat feel to the meeting, right? But I mean, you know, Vice President Vance said the same thing. I mean, like, there was agreement that we should have people that are thinking about more than like how the technology is effective, but think about it more from like a humanitarian impact perspective, that I thought was really interesting to that third-party auditor.