Failfinder70

An­thropic Is Tak­ing AI Welfare Se­ri­ously. I’m Not Sure It Knows What It’s Mea­sur­ing.

• Failfinder70 10 Jun 2026 14:29 UTC

  1 point

  0

  in reply to: Jonathan_Graehl’s comment on: Con­tex­tual Iden­tity Laun­der­ing: How Claude’s Image Re­fusal Can Be Routed Through Web Search

  I already tried submitted it to Anthropic, they ignored me, lol. Interesting thought on the politics, but running a test like that, I’d get stuck at the operationalization of “right wing”. The best i could do would be the old-school small l liberal vs small c conservative, but that doesn’t apply anymore...

An LLM Flagged My Paper About LLMs Flag­ging Things.

When Eval­u­a­tion Fails

Au­dit Re­port: Style over Sub­stance at Scale.

Con­tex­tual Iden­tity Laun­der­ing: How Claude’s Image Re­fusal Can Be Routed Through Web Search

Can Large Lan­guage Models Iden­tify Novel Threats? Part 1: Mir­ror Life and the Clas­sifi­ca­tion Gap