I'm excited to introduce our first speakers. Please welcome Aaron Levy, co-founder and CEO of Box, and Jeetendra Patel, president and chief product officer of Cisco. I think I see them.
Hey guys, how you doing? Great to see you.
All right. All right. What's up? What is up? It's great to see both of you. Thank you so much for doing this.
Good to see you. It was hard to take a couple minutes away from Fable, though. So, I'm surprised we got all these people in one room. And like, there's like tokens to be spending right now.
Did you give it a task while you were here?
Right. Yeah. We got to check in on it in a couple minutes.
I really hope everyone has their agents in loops right now, because that would be really inefficient if you don't. So yes, actually, let's start with Fable. That's a good...
I have a question. Does Max use AI now? Max is the son.
Max is my son. We're working on it. So, okay.
These guys, they know each other. They used to work together. That's what's going on here. So, we'll try not to...
Which I love. I love the banter.
He actually told me when I left, he's like, 'It's so great you left. I'd never work with you ever again.'
Only because at 2 in the morning, we would be like debating existential strategy. So it was like, it's nice to get a little bit of sleep, right?
Well, now you have your agents doing that for you. But speaking of Fable...
This is super interesting. So there was a huge backlash today on Fable from AI researchers who use AI coding agents to build their software, which they use to train open-source models, which then of course compete with Anthropic and other frontier model companies' closed-source models. And the allegation against Anthropic is that they are secretly sabotaging that code. The model can tell that they're doing that.
Well, it's actually not even secret.
But not even secret. But you don't know how it works, right? They're just noticing the output, right? And complaining about it. So I'd love to know, are you concerned that there's going to be anti-competitive behavior from these companies in the guise of AI safety?
I think it's... I would far prefer that they didn't sort of mitigate that in the model, because obviously, like, just AI progress is obviously... we want as much of that as possible. But I also, if you just understand the DNA of Anthropic, you can totally understand why they would do that. And it's an anti-competitive... you know, it's tough to use that term because it's their product. And so they sort of can choose how their product can be used to weaponize against them. Like, I think that's kind of like a reasonable, you know, competitive dynamic if you're using somebody's product to actually compete with them. Like, you know, that's sort of probably up to the core product to decide how that gets leveraged. And the great thing is, I think this is such a competitive market at this point that, you know, you have OpenAI that can easily not do that, and then that would become the product that more researchers end up using. So I think these things can be worked out through just the competitive forces of the market. So like, I love the complaints that researchers have, and I also understand Anthropic's position, and I think these things kind of work themselves out in just the battle of the actual market.
Before you became president, you were running security products for Cisco. You're also still...
Still am. Still am, right? Still in the purview.
You're also a big proponent of open source. So how do you feel about what's happening in that space?
I think like America needs to have a very strong open source strategy. And in the absence of that, even if there's no technical limitation of using a Kimi model within your environment, I think there's going to be a certain level of restrictions, and you need to have a good strong investment in open source in addition to proprietary models that are out there. And I think NVIDIA is doing a good job right now with NeMo and the entire portfolio that they're investing in. And by the way, in the past like 96 hours, a lot happened in that entire domain. So like, a new image model came out, a new video model came out, new language models came out. I do think that that's going to be a self-propelling thing because there's going to be enough demand signal. Now what's happening is the costs of tokens are far higher than the actual value that these tokens are generating at scale. And so what ends up happening, and the big risk in the market is, if you don't create an equilibrium there, then people just pull back on using tokens, and that's actually not good for anyone. And so what you need to have is there'll be an intelligent routing layer. You'll need to make sure that you actually have small language models. Not every query needs to go through a frontier model with 10 trillion parameters. And as you redirect down to small language models, you'll also have open source that's being used. And so in my mind, I think the token economics will actually dictate that open source is a very, very important play, and it'll be something that will be good in the national interest of America, and it'll be good also for corporate customers to have choice.
I mean, Aaron says it's fine for Anthropic to sort of defend their turf, but we know that's not what they're going to say if they get a call from regulators, right? They're going to say, 'We're doing this because it's safe. We're worried about AI safety, and this model's too dangerous. It's going to create a bunch of security flaws.' Are you worried? Now, Mythos obviously already sort of started this debate, but like, do you actually buy that? Is it that much of a step change in security if they allow this to run free?
We were in Project Glasswing. We were the first of the companies to actually get access to...
Glasswing is the Mythos program.
That's the Mythos program that had 11 companies initially that they made the model available without guardrails to. And so we've been... our product teams have been working very hard to make sure that you can actually patch the vulnerabilities and all that before Fable came out. And so I actually feel like there is a massive step function of vulnerabilities that will be found. I think the time from when a vulnerability is announced to when an exploit happens is now a matter of minutes. Patching is hard. Patching takes about 45 days to go out and patch, and only 20% of the vulnerabilities that get found ever get patched. 80% go unaddressed because you just don't have enough resources. And so what you have to do is... and so the way that we see this happening is you're going to have to go out and patch the infrastructure, but you'll also have to have mechanisms in place for one, addressing vulnerabilities at the root, which means that go out and use memory-safe languages, because the majority of the vulnerabilities happen with C and C++ kind of languages where you might have attacks that happen because of manipulation of memory. And so that'll be one area. And then the other area is what can you do during that window of time when you're exposed because a patch has not occurred because you have to get your systems down. So can you provide some kind of compensating control in the interim while the patch is not applied? And those are all the areas that we're working on. But it is a non-trivial risk that you should not take lightly, and it's a national security risk if some of these critical infrastructure... I mean, when you have a power plant or a hospital go down, and you can do that in coordination with agents, like...
