Comment by Animats

We need a way to make tight little specialist models that don't hallucinate and reliably report when they don't know. Trying to cram all of the web into a LLM is a dead end.