Takeoff math: separate training compute from the compute to actually explore
1a3orn · x · 2026-11-22
1a3orn breaks down the compute accounting for AI takeoff: you must distinguish the compute cost of training a model that does algorithmic improvements from the cost of the trained model actually exploring for more efficient discoveries. The same split applies to math models — cost per task when training vs. when proving RH.
More from AGI Musings
- Ex-NSA's Perlroth: AI is multiplying cyber risk, exploits now beat patches by 7 days — chrisrohlf · 2026-11-22
- Szegedy predicts proofs without formal verification won't count within a year — aran_nayebi · 2026-11-22
- Quant firms 'print money' with reasoning models — and early access looks like cheating — scaling01 · 2026-11-22
- Two compute costs that decide AI takeoff: training the improver and paying for its exploration — 1a3orn · 2026-11-22
- Largest Lean proofs cost at most ~$50 in compute, per Claude's estimates — 1a3orn · 2026-11-22
- a16z: LLMs update predictions in mathematically precise ways — but AGI needs continual learning — vishalmisra · 2026-11-22