AI eval diversity debate: the real fix is a market, not ideology
joshua_saxe · x · 2026-09-13
- Original claim (adibaradwaj): METR is the closest thing to an independent standards body for AI and an obvious evaluator choice, but its staff largely share the x-risk views of OpenAI/Anthropic leadership; labs should let third-party evaluators with diverse viewpoints, including voices outside the EA/rationalist movement.
- Counterpoint (MackenZarnold): This misdiagnoses the problem. Like arguing food pantries discriminate against atheists because churches run them — pantries are run by churches only because morally motivated actors are the ones willing to enter that market.
- Core argument: The lack of evaluator diversity isn't ideological anti-selection; it's that making a living in AI evaluation currently requires philanthropic money. The fix is building a market where more diverse actors can sustainably operate, not blaming incumbents' values.
More from Safety
- Blogger: Dario's AI safety essay reads like US-style military-civil fusion in disguise — bilawalsidhu · 2026-09-13
- e/acc founder mocks frontier safety evals as friends grading friends — beffjezos · 2026-09-13
- Chamath's warning goes viral: 'zero data retention' is no guarantee — run open models yourself — demian_ai · 2026-09-13
- tszzl predicts open-source models will be banned after a major AI disaster; Beff pushes back — beffjezos · 2026-09-13
- Alignment Won't Fix AI Security Threats — Hardening the Internet Will, Researcher Argues — nptacek · 2026-09-13
- Is scope creep worsening as models improve? Questioning the evidence behind HF incident takes — kuza55 · 2026-09-13