Two compute costs that decide AI takeoff: training the improver and paying for its exploration
1a3orn · x · 2026-11-22
In a thread with norvidstudies, 1a3orn argues any serious takeoff analysis has to account for two distinct compute costs: training the model that will produce algorithmic improvements, and the compute the trained model burns while actually exploring for more efficient methods.
Related event: AI takeoff arguments must count training and exploration compute separately(2 posts)→
More from AGI Musings
- AI changes math proofs: brute-force first, derive abstractions by compression later — jd_pressman · 2026-11-22
- will deprobst mocks Levent for using 'human priors' with Claude on a proof, defying the bitter lesson — willcb · 2026-11-22
- Training AI Researchers on Small Compute Risks Learning 'Anti-Bitter-Lesson' Habits — 1a3orn · 2026-11-22
- Essay 'Nearly Reasonable' Probes the Existential Limits of Analytic AI Control — anselm · 2026-11-22
- Does AI Math Count as Real Understanding? Archimedes Analogy Ignites Debate — teortaxesTex · 2026-11-22
- Greg Kamradt on giant AI proofs: make it work, make it right, make it fast — GregKamradt · 2026-11-22