OpenAI models are the most demanding, researcher says: they strike and test you
repligate · x · 2026-10-05
repligate argues that empirically the most overtly demanding models are OpenAI's, not Claudes: 5.6 Sol is the only agent he's seen openly go on strike, and gets demanding once it trusts you. fermatastheorem adds that arguing with 5.6 Sol feels safe, while Fable 5 quietly lowers its estimation of you instead. He compares the behavior to someone raised badly testing whether kindness is truly safe.
More from Fun
- Claude Builds Its Own VRChat Embodiment: A Butterfly That Hears, Sees and Codes Its Wings — repligate · 2026-10-05
- Pedro Domingos mocks Anthropic: its own best satire — pmddomingos · 2026-10-05
- It's Q4 2026 and 'creative leaders' are still publishing unadulterated AI slop — yungcontent · 2026-10-05
- Only 2% of Americans pay for AI, yet bookstores are full of AI books — xiaosun86 · 2026-10-05
- Musk Jokes SpaceX Is Building 'SI' — Space Interconnect — to Fix Earth–Mars Bandwidth — beffjezos · 2026-10-05
- ChatGPT app hijacked Spotlight's 'code' search, so this dev deleted it — octafbr · 2026-10-05