I can't share a lot at this point, but clearly how much we we allow the model to think is the main axis of the simplification.

→ View original post on X — @lmthang