Machine-assisted search finds a counterexample to a 87-year-old conjecture
MoonL88537 · x · 2026-07-20
The post highlights a math result discovered with machine-assisted search and argues it may signal a broader shift in how difficult problems are found, even if it is not a full paradigm change.
From the quoted discussion:
- An 87-year-old conjecture appears to have been disproved by a compact counterexample in three variables.
- The discovery process seems to have involved a workflow of semantic generator → structured candidate → exact verifier → revision.
- The key point is not blind hillclimbing, but using a learned proposal distribution plus a perfect verifier to search a huge space efficiently.
The author’s take is that models may not need to become omniscient theorem provers. They may only need to generate high-value, machine-checkable candidates from enormous search spaces, where generation is hard but verification is cheap.
More from Research
- Token Collapse: Study Finds Graph Soft Prompting Tokens in LLMs Fail to Encode Structure — Sapienza · 2026-10-27
- Terence Tao's SAIR Foundation revamps Mathathon with $1.5M compute per team — _sathvikr · 2026-10-27
- Mathathon 2.0 launches with Terence Tao's SAIR, $1.5M in compute for teams — _sathvikr · 2026-10-27
- The 'sweet lesson' is real: two kinds of ML problems need opposite strategies — mike64_t · 2026-10-27
- PixGR: Pixel-space generative decoding for fast feed-forward 3D reconstruction — kwangmoo_yi · 2026-10-27
- UniAction: unified action conditioning lets one world model handle mixed action types — ZeYanjie · 2026-10-27