Why 'LLMs are just next-token predictors' is a contentless argument

Aaroth · x · 2026-09-25

Responding to the recurring Twitter/Bluesky debate over whether LLMs are "only" next-token predictors or stochastic parrots, Aaroth argues the label is technically contentless: any mapping from inputs to output strings can be factored into a sequence of next-token distributions. It's a boring syntactic fact about probability distributions, and makes no claim about whether brains work like LLMs internally—so using it as a put-down isn't a substantive argument.

Original post →

More from AGI Musings

AGI Musings channel →