We desperately need better ways of evaluating models. Something that shows how helpful they are at working hand-in-hand with humans to help them get stuff done in a cooperative/iterative way. The Claude models have consistently been better at this, and the market rewards that.

→ View original post on X — @jeremyphoward