I'm not convinced LLMs can catch interesting deep patterns in their training data

→ View original post on X — @simonw