Beth Lyons

So like, time to move.

Yeah, there was a lot of discussion at the conference

Well, a lot.

There was some discussion in the conference talking about Scaled actually being, but Scaled's not Brian Scaled, but Scaled that is the synthetic data creator being part of why, even though...

The answers that LLMs give are kind of similar across the board.

One of the reasons for that was that there was a corpus of data that all teams had.

Called the internet.

And maybe yes, maybe no, that will be a problem.

But they were all training on like this big chunk of data that included books and a variety of things.

Yes, called the internet.

And so that created a specific set of probabilities, right?

Well, they moved to a lot of them also using their data from SCALE, right?

So we haven't really diversified from what the training stuff is, so they're still giving kind of similar data.

There's actually a great explanation, I'll bring it tomorrow, that I saw on X about how to look at that in terms of what a weight means,

Well, where the data comes in, that kind of stuff.

Used a baking analogy.

Great.

So my next, let's see, where am I pulling this from?

Andy, give it a short and I will pull mine up.

Okay.