LLMs are huge piles of neurons that somehow give useful outputs, but at which points do real concepts emerge from this mathematical mess? @Anthropic team did fascinating work on that: read my summary here

→ View original post on X — @aymericroucher