Stop Thinking of LLMs as Next-Token Predictors (via square_usual) — discussion

#vibecoding