Community LLM auto-research boosts integer multiplication result 69% in 24 hours
RexDouglass · x · 2026-10-12
A community auto-research experiment combining LLMs with Lean formal verification pushed integer multiplication bounds from κ=2⁻¹⁸² to 0.001289283116, a +69.13% jump in 24 hours.
- After an open call, multiple experts submitted new designs; @mrowdays drove most of the gain and introduced Sussman's E8 geometry (a symmetric lattice in 8 dimensions), adapted by contributors on both the complex and bit sides.
- Observers see a new kind of academia emerging: anyone with spare LLM quota can contribute, Lean replaces peer review, work is orchestrated in anonymous group chats, progress is much faster but incremental, and agent legibility now matters more than human legibility.
More from AGI Musings
- Jeff Holden on microscale manufacturing: giving compute a micron-scale body — anselm · 2026-10-12
- eigenrobot uses ChatGPT to audit an economist's CV: all 20 papers co-authored, zero solo — eigenrobot · 2026-10-12
- Weekly AI Reading: Nadella on Models as Insider Risks, TSMC Revenue Up 54.6% YoY — VibeMarketer_ · 2026-10-12
- Reddit debate: Should AI agents specialize and trade services to stop wasting duplicated tokens? — blafri · 2026-10-12
- Is AI coming for the g-loaded and verifiable disciplines next? A hypothesis and a caveat — morqon · 2026-10-12
- Alexandr Wang: billions with AI agents will switch on everyone's agency at once — davidsenra · 2026-10-12