Researchers push back on Dwarkesh: defining AGI metrics is the hard part
dhadfieldmenell · x · 2026-11-15
Cindy Wu bets against Dwarkesh Patel's AGI optimism: climbing a metric requires defining it first, and defining good metrics that resist Goodharting and reward hacking is most of the game of making AGI. Data and environments do the lifting—'AI will just automate data' elides the hard problem—and ablation over N changes carries naively exponential compute cost. Hadfield-Menell endorsed the thread.
More from AGI Musings
- Does 'Meditations on Moloch' Imply a Friendly AI Singleton Is Our Best Bet? — jessi_cata · 2026-11-15
- A barber only heard of ChatGPT last week: real-world AI adoption gap — AI_Andrew · 2026-11-15
- AI researcher: nobody can tell X replies are bot-written — and that's scary — iScienceLuvr · 2026-11-15
- Pedro Domingos mocks AGI hype: AI researchers never feel AGI is near — pmddomingos · 2026-11-15
- Pedro Domingos asks: how much false info is in your brain from trusting LLMs? — pmddomingos · 2026-11-15
- Blogger signs petition to reinstate Jensen Huang, citing his open-weight advocacy — esrtweet · 2026-11-15