Imo the biggest open research area in LLMs now is how to exactly do evaluation correctly. It's tricky as pointed out below

→ View original post on X — @yitayml