Emergence debunked? Researcher argues emergent LLM abilities are labels in your imagination

gerardsans · x · 2026-09-29

In a thread responding to a timeline of emergent capabilities, data scientist Gerard Sans argues that none of the listed cases prove emergence, pointing to a paper from roughly two years ago (the "mirage" argument) that debunked it. He likens it to primes not being emergence in natural numbers: labelling model behaviors is an external mechanism living in the observer's imagination, not in the training data, the transformer, or inference.

Related event: Debate Reignites Over Whether LLM Emergent Abilities Are Real or Illusory(3 posts)→

Original post →

More from AGI Musings

AGI Musings channel →