Why 1k-10k agents per strong math solution may be the right order of magnitude
cephaloform · x · 2026-09-09
Commenting on OpenAI's math proof announcement, a user argues the reported 1,000-10,000 agent scale makes sense for reaching a strong solution — or alternatively, using stack-based languages with a tiny 10M-parameter AR model vomiting equations, sampling at least 512 times per step.
Related event: Community Weighs In: Solving Hard Problems Needs 1K–10K Agents(2 posts)→
More from AGI Musings
- Turing winner Boaz Barak: an AI solved a Millennium Problem and there's plenty of headroom left — soumitrashukla9 · 2026-09-09
- An Unreleased AI System Solved a Hard Puzzle in Under a Week, Igniting Concerns Over Concentrated Power — abhiadesai · 2026-09-09
- CSET's Helen Toner: the term AGI is becoming useless as different camps mean entirely different tipping points — hlntnr · 2026-09-09
- OpenAI says 10,000 coordinating AI agents solved Navier-Stokes in 88 hours — CatAstro_Piyush · 2026-09-09
- $22.5M of Compute in One Week: The New Price Tag for Math Breakthroughs — vykthur · 2026-09-09
- Navier-Stokes solve points to labs quietly reserving frontier capability, says dev — idanbeck · 2026-09-09