The inference process of current LLMs is somewhat comparable to attempting to answer a question without deep analysis, ie through a process of pattern matching, creative association and constraint resolution. However, reasoning is itself a process of successive pattern matching.

→ View original post on X — @plinz,