Joe Liemandt

And that's a critical aspect of it.

That's why if you just have, and why I believe all the LLMs that are out and all the educational ones, they're not embedded with learning science.

They don't follow these principles.

They don't embed mastery and all these other concepts.

And that's why generally when you give them to kids, they just use them cheap.

So you have that side.

Then the second side is you're like, okay, I have this data stream and now I'm basically, it's a video model.

It's just a vision model.

And you're just sitting there saying, how do I optimize this vision model to find what I need?

But how it started is literally just

Humans, when you talk about human annotation, just watching videos.

So three years ago, we would just sit there and have teams of people all night.

You know, the kids would finish.

And then all night, you would just sit and watch the videos.

and annotate them of, okay, that's what's going wrong here.

This is what the kid's doing.

All of this.

It's like reinforcement learning.

And it is.

And it's just building out a video model, a reinforcement learning video model with human feedback.