Can LLMs Out-Patience Mathematicians on Collatz? A Thread on AI-Driven Proof Search
an_interstice · x · 2026-09-14
jdpressman and aninterstice discuss using AI (an agent called "Astra") to attack Collatz-style math problems:
- Core claim: such problems are full of grind-y blind alleys and nested obstructions where human mathematicians go in circles (like the infinite staircase in SM64); an LLM's "infinite patience" could navigate the maze to its center.
- Evidence: when Astra searches literature about a given alley, it usually finds relevant material, often directly about Collatz—implying rough consensus on which questions matter, so the problem's structure is known and only a resolving pattern is missing.
- Current state: on the tricky parts Astra nibbles out partial results, but each creates new obstacles; presumably at some point a base case is reached and the whole stack of obstructions rolls up.
More from Research
- float32 evaluation can break cryptanalytic neural-network parameter extraction attacks — chaumian · 2026-09-14
- Style-aware paraphrasing cuts authorship attribution F1 by 60-70% while preserving meaning — chaumian · 2026-09-14
- RL training in practice: "batch size 4" really means ~80 trajectories per step — cephaloform · 2026-09-14
- FlyOCR: A Simulated Fruit-Fly Brain Reads PDFs at 87% Character Accuracy — llama_index · 2026-09-14
- 11-page paper shows how group-averaging turns torpid MCMC mixing rapid on Ising models — michaelchchoi · 2026-09-14
- FlashREINFORCE: first open-source critic-free single-rollout async LLM RL with 6,000+ stable updates — YouJiacheng · 2026-09-14