The sampler is not the data: untangling system, process and output in LLMs

gerardsans · x · 2026-09-29

In a debate with @morallawwithin, gerardsans systematically rebuts the claim that 'a model is just a sampler, hence no intelligence.' His example: a model outputs General Relativity and you read intelligence from it; swap the data for cat pixels and the output changes to a cat while the sampler never changes. He argues the critic conflates traits of the system, the process and the output.

Related event: Debate Flares Over Whether LLMs Are Just Samplers Without Intelligence(2 posts)→

Original post →

More from AGI Musings

AGI Musings channel →