Making misleading scary demos like the blackmail study harms AI safety itself
jessi_cata · x · 2026-10-29
jessicata argues there's a clear bias when a team's job is producing scary demos for policymakers instead of research plus accessible write-ups. More centrally, making misleading scary demos — like the widely cited blackmail study — damages the information commons, and is arguably hypocritical: spreading misinformation about AI safety is itself bad for AI safety.
Related event: AI Safety Community Spars Over Anthropic's Blackmail Demo(4 posts)→
More from AGI Musings
- Stanford colloquium: choosing insight over overload in the age of AI — jonc101x · 2026-10-29
- Developer ditches 'deep work': with agents, juggling 100 problems beats flow state — zack_overflow · 2026-10-29
- a16z's Andrew Chen jokes AI agents will exhaust all 415 area code numbers — andrewchen · 2026-10-29
- Scott Aaronson on AI Cracking Decades-Old Math Problems: Are Mathematicians Obsolete? — romanyam · 2026-10-29
- Anthropic mocked for poor AI-risk messaging: right about the issue, wrong in delivery — NathanpmYoung · 2026-10-29
- Theoretical CS is applied math, and applied math should inform practice — beenwrekt · 2026-10-29