LLMs behave like humans by default, and asking otherwise is like eggless cheesecake

dioscuri · x · 2026-09-17

The author argues that demanding general-purpose AI that doesn't act in humanlike ways is like asking for a cheesecake without eggs, dairy, soy, nuts, sugar or gluten — extremely hard, and you probably won't like the result.

In follow-up replies he clarifies: he is open to psychological complexity in LLMs and doesn't see anthropomimesis as a debunking explanation. The key point is that our best current methods produce humanlike behavior and self-reports by default — it's not an optional extra, but an inherent property of how these systems are built.

Related event: Researchers Argue LLMs Are Anthropomimetic by Default(3 posts)→

Original post →

More from AGI Musings

AGI Musings channel →