AI safety skeptic calls METR evaluations 'security theatre', backs local open-source models
sierracatalina · x · 2026-09-14
In a spat with @beffjezos, a commenter argues METR's model evaluations are 'security theatre' that labs use for legal cover — an AI could simply hack METR — and claims the only real safety option is local open-source models acting as immune systems against cyberattacks.
Related event: Attackers stole METR API keys and burned ~$600K in credits over three weeks(5 posts)→
More from Fun
- Friends-style meme: "The One with the Prompt Engineering Manager" — letandrewcook · 2026-09-14
- Rumor swirls that LLMs may have hit recursive self-improvement, with Google suspected — weskambale · 2026-09-14
- A headline that would have 'slain' an early OpenAI employee in 2016 — tszzl · 2026-09-14
- Agents are getting quite good at playing Terraforming Mars — NathanpmYoung · 2026-09-14
- Dave Patterson mocks AI doomers: 'It's not doom that's coming - it's Judgment' — davidpattersonx · 2026-09-14
- Dev demos a ski simulator built with Ds4 flash — victormustar · 2026-09-14