Grant Harvey

Excellent.

So as far as the benefits for open source AI for the end user, how would you sum up those benefits for people who are used to using ChatGPT but maybe have never downloaded a model, run it on their own computer, or run it over the cloud?

And then, so I guess specifically with what you're doing at Reflection, you know, one, what are you building specifically?

And two, how do you plan to compete with the big labs with what you're doing?

So for people who don't know, so we've covered mixture of experts, which correct me if this is wrong, but essentially it's like a lot of tiny models that are highly specialized, like the experts in a specific field of interest.

And then you have the dense model, which that would be like what we think of as like a large language model, like GPT-4 or 4.5 or something like that, where it's just, it's trying to remember everything at once.

Is that correct?

Okay, let me give you a, because this is super fascinating.

I love learning this.

So let's say for sake of argument, I have a prompt and my prompt is, you know, write, this is like a classic one Corey does, write about how to make spaghetti, but from the point of view of a quantum physicist.

The path that it would take would then be like, okay, recipes and cooking, but then also quantum physics and then like pulling in all of these things as they're needed.

Is that kind of how it would work?

I like that.

I'm glad that we ironed that out because sometimes it's easy to think of it in a higher level of abstraction, but you're saying this is happening at the weight level.

Yes, exactly.

Yeah, gotcha.

Okay.

Well, I mean, speaking of weights, you're releasing the model weights for the open source models that you're creating.

Walk us through what you're releasing and why that's meaningful for developers.

When we talk about environments, are we talking about just code environments or almost simulating actual websites and things like that, since we're talking about an agent here?