Did you basically build your own model or you took your data and made it a proprietary data sitting on a model? How did you do that? This is for your colleagues who are in similar situations.

Sure.

We experimented with our own proprietary models in the early days.

I think at this point, we use a variety of models, both closed weights, open weights, various sizes.

I think...

The model conversation, if you look kind of at the market, I don't want to say it's irrelevant, but I think it's increasingly becoming a red herring, right? If you look at the frontier models, they're all fairly close to each other.

They all have their peculiarities.

What is, just briefly, what is your definition of, let's just say responsible AI?