Pinned
I'm tired of the "LLMs don't think and just statistically predict next token", so here's a quick overview of the common theses:
"LLMs are just statistics". No, statistics doesn't include multi-layer models with emergent intermediate features and non-linear dependencies.
