AI conjecture hunting could clear out many solvable math problems first
anderssandberg · x · 2026-07-26
- Anders Sandberg says it is fun to feed conjectures to AI, and shares a geometry problem that has resisted resolution so far: whether there exist convex polyhedra with no non-overlapping unfoldings.
- He notes that the Barvinok–Ghomi pseudo-edge construction comes close to disproving it, but Sol could not turn it into a true edge construction.
- The thread frames AI as a tool for sweeping away conjectures that existing methods could have solved, leaving the genuinely new-math cases behind.
Related event: Sandberg: AI Will Cull Solvable Conjectures First(3 posts)→
More from AGI Musings
- Elon Musk says civilizations die when they lose faith in the future — elonmusk · 2026-07-26
- “Coding is 90–95% solved,” but math still looks open-ended — haider1 · 2026-07-26
- Zachary Lipton backs making model distillation fair game for US companies — zacharylipton · 2026-07-26
- Chamath says open-source AI cannot be dumped, and China’s real edge is electricity — joshua_saxe · 2026-07-26
- Schmidhuber backs open-source model distillation and cites his 1991 work — SchmidhuberAI · 2026-07-26
- Programming Languages Becoming Computer-to-Computer Communication, Researcher Notes — eptwts · 2026-07-26