The Exchanges

Q look onto that. I've, I've heard rumors about, you know, the AI legal space, and there might be huge high-flying companies that rhyme with shmarvy, but, you know, not many actual lawyers are, are using it in-house. So, like, implementation inside is, is low, but adoption is really spread across. How deep is AI, like, actually implemented within the enterprise, and How long do you think that's going to take?

A Yeah. So we're transitioning from this phase of everyone for a couple of years have known that they must say to the world that they're using AI for everything. Every board will have said, like, we have an AI mandate. We're going to adopt or die. And so that means everyone talking a big game about how they're using AI. In practice, there have been a lot of proof of concepts, sandbox, enterprise deployments, internal deployments, kind of in low stakes scenarios. And that's enough that you can say you're using AI. But there's still a gap for this to, like, when you look at what is the number of lines of code that are being contributed to JP Morgan's code base that's generated by AI, as of very recently, that's extremely low. It's now starting to happen. We're starting to make, like, our way into, like, real use cases, both hospitals adopting this, banks adopting this, and businesses adopting this. But it feels like it's really in the last six months that you're moving from toy projects to kind of real adoption. But still that's just the frontier companies that are, uh, Like when you look at the long tail of businesses, they're still like, uh, still lots of the world runs on paper, right? Like there's still a long way to go.

AI assessment note: “moving from toy projects to kind of real adoption. But still... a long way to go.”

Q I'm not sure what the range is. It might have been, I'm not sure. We can check. But, uh, could you break down what is a kill switch in AI?

A Yeah. So a kill switch is like, does anyone anywhere have a button that they can press if something really bad happens that shuts off AI? And this is really interesting when one of the concerns that's coming up now that AI is getting really good and getting really agentic is that there might be what's called a loss of control scenario where Basically the AI agent becomes so agentic that it starts to take action that we no longer want. It might copy itself onto servers that we can no longer control. And all of a sudden humanity stands with a big disaster on our hands. And so a close kill switch is like, can we in all deployment scenarios, retain a button that big red button we can press to make this go away. And maybe an analogy is like in any given building, you're going to run a, Uh, a fire, what's it called? A fire drill, where everyone exits the building. You cannot run a company where people can't exit the building pretty fast if something really bad happens, and that's the same thing here. Um, whether that's going to happen by any of you, I don't think so, but it's, um, it's growing an awareness, even in, in this administration, you see, uh, David Sachs articulating like these kind of catastrophic scenarios. Uh, not outside of the realm of possibility, and so you need to plan for it. You actually also hear Chinese government taking, like, using the term kill switch pretty …

AI assessment note: “a kill switch is like, does anyone anywhere have a button that they can press”

Q VCs DM me on X, and I can get all your portfolio companies set up with a free samples credit deal. Your investors are super ingrained into the space, especially not Freeman. I mean, he's now a part of the super intelligence team at meta. What kind of advisement are you getting from them? How close are you, um, And like, what is the relationship between you and your investors?

A Yeah, um, we're really close. They're just, like, Nat is just the, I would say, the preeminent AI investor. Both made a lot of great investments, but also, as you just mentioned, is very close to the metal as to how this is actually being built. Uh, both the small questions of how to build AI, but also some of the big questions. Um, Nat himself spent a lot of time on Some of the risk questions that we were dealing with when he was leading GitHub, they built GoPilot, which is like the first real AI product that worked, and he led the efforts that led to Microsoft basically ensuring all of their customers against IP risk to say, oh, they're really worried about IP. We're just going to step in, and if they get sued, we're going to cover their ass.

AI assessment note: “we're really close. They're just, like, Nat is just the, I would say”

Q Again, super. And then with Anthropic, maybe we could start with Anthropic and then go to SSI. They have different levels of safety. So what do those levels mean? Uh, and could you break those down?

A Yeah. So they call them security levels, security level one, two, three. Really it's a process of saying when models get sufficiently good and therefore sufficiently dangerous, we need to match that with the levels of security we're implementing. What are the level of tests we're going to run? What is the level of Cybersecurity protection we need to have in place. What are the kind of policies that match each of these security levels? And so we're slowly climbing through the temperatures now. Ananthropic, um, you were talking about just before we got on here about whether security level three is going to be a thing. And I think the short answer is, yeah, it's just a question of time. And is it going to be this side of Christmas or after? I don't know yet, but it's really like company's ways of matching proportionally the risk with the mitigations.

AI assessment note: “Really it's a process of saying when models get sufficiently good and therefore sufficiently dangerous”

Q How do you go down to the code base for this? Because Things are moving so fast. Like, how do you model this out?

A Core of the standard is we actually just have to test how the agent, AI agents perform in real life settings. So if you're worried about, if you want to know whether an AI agent can get jailbreaking, broken, the best way to figure out is to try and jailbreak it. And the field is moving really fast. So what's really important is that you build this connection between the latest research papers and what the standard It does, and the kind of testing that's required. So where some of the old standards get updated every five, 10 years, we're gonna update our standard, AIUCA-one, every three months to make sure that as new kind of threat vectors come out, as new capabilities come out, that the testing that's being done reflects what's possible.

AI assessment note: “we're gonna update our standard, AIUCA-one, every three months”

Q Do you think that trend of just spending a ton of money on go to market is going to continue? Like, how does that evolve as things advance?

A Yeah, I think we're in a super cycle. I think we are. It is still, it is wild to think this, but it's still early days and, um, I think people are still losing their minds over their salaries that Meta is willing to pay, but it is a direct consequence of taking this worldview really seriously, that in 10 years everything looks unrecognizable, and that whoever's on top now may not be on top in a year's time. That is what's predicating this enormous amount of investment, both into the technology, but also into go to market. The question is really like, you can buy more compute. But it's hard to buy your customers confidence. You really need someone else to say that you're trustworthy. Like, no matter how many billboards you put up saying like, we're really the good guys. We're really responsible. In practice, it just turns out that in most industries, having third parties go and look at that is going to be more effective. Um, so hopefully we can find ways for them to like actually earn the confidence, not just get a stamp on it, but actually do the things that means that they deserve the confidence of their customers.

AI assessment note: “That is what's predicating this enormous amount of investment, both into the technology, but also into go to market.”