"Next-token predictor" is the wrong mental model for LLMsgmcgoldr.github.io
24 points by garrinm 4 hours ago | 68 comments