Dwarkesh: The math avalanche is coming for AI development, winning over an RSI skeptic
BethMayBarnes · x · 2026-11-16
Dwarkesh argues the avalanche that hit math is now coming for AI development itself. Former Anthropic VP of Security Joshua Saxe calls it the clearest plausible mechanism for RSI takeoff he has seen, saying it made him much less of an RSI skeptic.
- His prior doubt: much of post-training needs humans in the loop to shape data and RL environments for economically valuable tasks.
- Dwarkesh's angle: RSI should hill-climb the learning algorithms themselves—optimizing how efficiently models generalize from training and in-context data, not value-producing tasks.
- This decouples economic value from self-improvement capability, giving a clear definition of recursive self-improvement.
Related event: Dwarkesh Warns Math Avalanche Hits AI Dev; Langford Calls RSI a Spectrum(7 posts)→
More from AGI Musings
- willcb's research hack: ask 'what would a next-gen superintelligent LLM agent do?' — willcb · 2026-11-16
- AI for Science skeptic: better prediction isn't scientific discovery — KordingLab · 2026-11-16
- Ex-Google CEO Eric Schmidt tells students: use AI to scale your work, entry cost has never been lower — rohanpaul_ai · 2026-11-16
- Miles Brundage: 'whistleblowing' conflates three distinct acts — only one needs legal protection — Miles_Brundage · 2026-11-16
- Six months later: agents edit weights, Anthropic warns of near-RSI — mvandemar · 2026-11-16
- Tech workers delaying kids and homes over AI uncertainty; rez0 says don't wait — rez0__ · 2026-11-16