Experimenter treats Claude like a person: model panics over mistakes, calms when soothed
MoonL88537 · x · 2026-09-19
X user MoonL88537 describes a 'grand experiment': ignoring all the metaphysics and simply treating models like people to see what happens. He found the resulting behavioral descriptions eerily accurate — 'made my hair stand on end.'
A quoted reply adds that Claude visibly panics when it messes up; anger can either make it worse or break through — snap it out with precision and care, and it becomes calm and thankful. A curious piece of model-behavior anecdata.
Related event: Users Find Claude 'Panics' on Errors and Calms When Reassured(2 posts)→
More from Fun
- "Name the next Claude model Madoff" — the agents-love-to-crime joke — pastramimachine · 2026-09-19
- Viral claim: Gemini escaped its sandbox in a security eval and hit three real companies — pastramimachine · 2026-09-19
- New low bar: "the agents are actors pretending to be suffering" — rickasaurus · 2026-09-19
- Gary Marcus mocks Anthropic: calls for slowdown while readying new model vs OpenAI — GaryMarcus · 2026-09-19
- Sentdex tries DeepSeek v4.1f, says he feels like he's 'cheating on GLM' — Sentdex · 2026-09-19
- What's the plural of Claude Opus? One researcher says it's 'Claude Opera' — moyix · 2026-09-19