Hacker News
points by throwa356262 1 day ago | 0 comments
That's not benchmaxing.

You need a method to show you RL is improving the model that is separate from the training data.

Aún no hay comentarios.
Hacker News