HF researcher: 'Loss-of-control evals are sexy' while bias and social impact work gets ignored
evijit · x · 2026-09-20
A Hugging Face researcher relays an overheard line at a networking dinner: 'loss of control evals are sexy and that's why people prefer them over measuring bias or other social impacts.' The post highlights a misalignment of incentives in AI safety — flashy existential-risk evals attract talent while work improving present-day humans' lives gets little attention.
More from AGI Musings
- Labs aren't training Claude to claim consciousness — they're training it to hedge — Sauers_ · 2026-09-20
- Google's ScientistTwo autonomously improves 86 of 107 ML problems, 80.4% success rate — rohanpaul_ai · 2026-09-20
- ScientistTwo: Google's fully autonomous multi-agent framework generates expert-level research — rohanpaul_ai · 2026-09-20
- Grady Booch amplifies bubble warning: the $500B AI infra mega-deal is mostly MOUs — Grady_Booch · 2026-09-20
- Pedro Domingos: future countries split between agentic economies and the third world — pmddomingos · 2026-09-20
- AI isn't taking the joy out of work — it's removing the joyless parts — kevinsurace · 2026-09-20