METR and the case for diverse viewpoints in AI safety evaluation
One post praises METR as the closest thing AI has to an independent standards body, while urging labs to include evaluators outside effective altruist and rationalist circles.
TLDR
A post strongly supports METR’s inclusion as an AI evaluator but claims its staff are largely ideologically aligned with OpenAI and Anthropic leadership on safety issues such as existential risk. It argues that labs should allow third-party evaluators with diverse viewpoints, including those outside the effective altruism and rationalist movements.
Combined views
25
1 Source, first seen ago