Can you give us an idea of how your own AI factory will work? You know, if if companies if you're going to give companies the ability to build their own proprietary systems that are safe, secure and highly customized and even cheaper, wouldn't that threaten the frontier model?

Well, I think what we're going to start to see is due to the increased competition, some of the margins around the model itself starting to come down.

But we have to remember that when enterprises are buying something, they're not just buying a model, they're buying a system.

And that system involves the hardware accelerators, the cluster management software, inference software, the harnesses, and they're buying that whole thing end-to-end.

And we just see margins getting distributed across that stack and intelligence providers providing end-to-end systems for them.

What we're seeing is insatiable demand for agentic inference.

We don't see that really stopping.

Where do you come down on the pace the frontier debate?

It's a finance company, it's a medical company, et

cetera.

Let's start with the first question.

Yes, this is what we're seeing demand from large enterprise, sovereign, public sector agencies is the notion of taking in workloads that used to be rented on a closed model uh and now standing them up on their own infrastructure so you might be uh you know when when it's a sovereign it's typically a large conglomerate within some allied international country and they you know are they want to get into the token business right they want to stand up uh a cluster so they'll get a partner to help them stand up the bare metal but then they want to serve tokens into their region right and the only way for them to serve tokens and basically become the intelligence provider for that region is to work with an open model by definition because that's the only thing that they can take and run on their stuff.

So we help those kinds of conglomerates.

We're working with them in a number of countries.

There's also just Fortune 500 Enterprise, and it tends to be somewhat similar there.

It's

Mm-hmm.

So the application, you know, startups like, uh, Cursor or Cognition or Harvey, uh, like these AI natives Abridge, uh, that initially started their businesses on closed models.

But as their businesses ramped up, reining in kind of control of the intelligence over multiple factors, over cost control by fine-tuning those models on their data and ensuring that they have the best cost for the capabilities that their products have.

Compute security, right? They can get compute and run the models, and they under-- and they know that that compute will always be available because it's their compute.

Data security, right? Ensuring that their data doesn't leave, uh, and is sort of, uh, sovereign to them.

Uh, there are a number of reasons why, you know, these startups, as they've gotten very big, they start adopting open models.

And so- Without an open model ecosystem, without capable open models, the most promising startups cannot really become independent, right?

Yeah.

What do you think happens over the next year or two in terms of Chinese open source or Chinese open weight models? Do you think it remains where it's at? Do you think it evolves? Like, what happens?

So I, I think the first thing is that the fact that great, a great open model ecosystem came from China is actually a massive benefit for the world.

Like, the, there are so many companies in the West, right, that have been able to build more durable businesses as a result of that.

So it's, uh, uh, there is a sense in which it's a, right, it's a very positive thing and a gift because you could imagine a world where such a thing didn't happen, and then there are just no open models, right? And there's a kind of a tall poppy syndrome that happens when if you're a application builder and you build something that's great, well, then all, it gets subsumed into, right, a closed model provider because, well, then you gen- it's a business.

It needs to generate revenue.

So I, I view it as positively.

Um, the thing that is...

How do you direct, like, the scientific and experimental effort now, like from, from Beam going forward? And, uh, within that, assuming that you are trying to use the models themselves to, um, improve your training effort, like what is the role of the science team today?

You bet on open in, what, this 2024? Is that right?

Well, I guess it was a year ago.

Well, actually, when we started the company, we had two bets.

And I would kind of bring it back to the origin story.

At the time, Yannis and I had worked on Gemini 1, 1.5. Yannis was the RLHF lead and one of the most storied reinforcement learning researchers at DeepMind.

I actually switched from physics AI when I saw AlphaGo come out.

And then I later learned, you know, well, I knew that Yannis was an author, but that was kind of a, you know, sort of a, dare I say it, a celebrity crush moment when I joined DeepMind and had the pleasure to work under Yannis on his RL team.

And then can you give us a sense of what's coming on the how do you get these models out to the world beyond just putting on hugging face or something?

Um, why are they involved?

So I think broadly, um, where we are today is that open intelligence matters both to America as a country and to large enterprises.

Uh, there is, um, a global AI race happening, both among existing AI companies, but also between the US and China.

And today the best open models, uh, are coming out of China.

And as a result, global AI adoption is actually, um, very much kind of being built on top of Chinese models, including American companies.

Um, and that's great, but we wish to build a pretty...

A, a thriving AI ecosystem that has a lot of competition and a lot of players, and we're building the counterbalance to that, building great open models-

Mythos is, for example, an incredibly powerful model by Anthropic said, "We're not going to release it because it would be too dangerous." Isn't there a danger to being open source?

But quickly, uh, give us a reintroduction to the company and, of course, give us the news.

Uh, a quick reintroduction to the company, uh, I'm Misha, I'm the co-founder and CEO of Reflection together with Yannis.

Uh, we started the company about a- a year and a half ago, and, uh, we were formerly of DeepMind.

Yannis was, uh, one of the founding engineers at DeepMind, uh, contributed to a lot of projects, uh, like AlphaGo and Gemini, and recruited a team of about, uh, 60, I would say, uh, researchers and engineers from Frontier Labs.

Uh, and the charter of the company, um, has opened up since we last spoke.

We are, we've raised this capital to really build out the Frontier Open Intelligence, um, in- based in America and exported to the rest of the world.

So-

Yeah.

Why don't you kick us off with an introduction on yourself and the company, and then I have a ton of questions about how engineers are, uh, understating code.

Yeah, sounds great.

So, um, just a quick primer on the company.

Um, my co-founder Janus and I were DeepMind researchers, uh, before this, um, or- ourselves and our team, which is mostly from DeepMind and also from places like OpenAI and Anthropic, pioneered a lot of the large language model and reinforcement learning breakthroughs over the last decade.

And we set out about a year and a half ago to build a company with the mission of, um...

I mean, it was with the mission of building superintelligence, but we had a pretty, let's say, almost practical and opinionated view on it that, like a not ability in the abstract, but you had to really define it as both the research and the product agenda.

And so we thought about, "Well, what would a superintelligence in an organization look like?" Which we think is basically gonna be this, like omniscient oracle that knows everything about the company and can act on the user's behalf.

Obviously, there's a lot of capital available for great companies and teams, but it's still quite a bit more resource constrained than something like a DeepMind.

And I'm curious what kind of reaction you've seen from your customer base with the notion that you want your tool to be able to do it on its own versus kind of have this human intervention alongside of it.

That's a really good question.

So actually, we've found that the tool, both internally at Reflection and the way our customers use it, is incredibly empowering because it allows engineers to focus on the stuff that is really high priority and they want to work on.

And we kind of take the more tedious work off their plates.

So we kind of think of an engineer of evolving over the next few years and becoming basically an architect that has an army of these hundred AI engineers working for them and helping them see their visions through.

I'm curious how you're going about the safety conversation a year afterward.

Yeah, that's also a really good question.

Safety is definitely one of the things we care about a lot.