Research & academia
“Next-token predictor” is the wrong mental model for LLMs
The captured discussion asks whether “next-token predictor” is a useful technical description, a misleadingly narrow mental model, or both.
Why it matters: The discussion cautions against letting a compact label do two jobs at once: describing token generation and explaining a system’s capabilities or limits. When evaluating a claim about LLMs, ask which meaning is intended and whose interpretation is being reported.