In a paper published last week, Apple researchers argue LRMs merely use a different kind of pattern matching without actually generalizing better than other LLMs — but don't necessarily improve performance or increase effort when doing more complex tasks.

→ View original post on X — @martyswant,