Researchers Argue LLMs Are Anthropomimetic by Default

DeepMind researcher dioscuri argues LLMs are anthropomimetic by default: base models naturally report inner lives, so their self-reports should neither be fully trusted nor seen as mere design artifacts. He likens demanding non-humanlike AI to baking a cheesecake without eggs, dairy, sugar or gluten.

2026-09-17 ~ 2026-09-17 · 3 related posts

• LLMs are anthropomimetic by default: self-reports of inner lives need scrutiny, says researcher — dioscuri · 2026-09-17
• dioscuri clarifies: LLMs are anthropomimetic by default, humanlike reports aren't optional — dioscuri · 2026-09-17
• LLMs behave like humans by default, and asking otherwise is like eggless cheesecake — dioscuri · 2026-09-17