Mathematician says AI agents gave him a decade's progress in one day, trains models for recurrent inference
nasqret · x · 2026-10-20
A math researcher reports that delegating work to AI gave him "more progress in one day than in an entire decade." His workflow: before reading anything, he lets agents bootstrap problems using custom skills and an agentic harness to eliminate dead ends (especially in algebra), then combines this with computer algebra into a self-correcting system producing genuinely novel reasoning. LLMs may be stochastic parrots at the bottom, but RL training builds much more on top. He also highlights SAIR's new Open Models for Mathematics initiative and his own experiments training models for recurrent inference against vanilla-trained baselines on his Heyting arithmetic library — inconclusive so far, but the models are learning. He predicts more mathematicians will use frontier models and calls for a community of mathematical builders.
More from AGI Musings
- Humanities dissertations have collapsed, social sciences are next — with AI as one driver — gsiemens · 2026-10-23
- Greg Brockman: AI makes hacking 100x more profitable and easier, upending hacker economics — gabriel1 · 2026-10-23
- 'Nothing is real now on social media': viral demo shows AI-generated content everywhere — saibharadwaj · 2026-10-23
- Lilian Weng to Discuss Recursive Self-Improvement at SGLang Summit 2026 — BanghuaZ · 2026-10-23
- mmitchell_ai questions whether the AI safety vs security framing stems from EA — mmitchell_ai · 2026-10-23
- AI, Robotics and Abundant Energy Will Undo Millennia of Scarcity-Driven Life, Argues Kardashev-Scale Optimist — Dr_Singularity · 2026-10-23