There's a lot more happening inside an LLM than just next token prediction

→ View original post on X — @alexalbert__