GPT Astra-assisted proof improves Erdős Problem #357 bound, Lean-verified but not refereed
kevrussell · x · 2026-11-03
Kevin Russell, with help from GPT Astra, reports a candidate step on Erdős Problem #357 (any order): for every k ≥ 10²², k distinct integers with all consecutive-block sums pairwise distinct must include numbers above (3/2 + 10⁻¹⁰)k, marginally beating Hegyvári's 1986 bound of 3/2.
- The argument has been formalized and checked in Lean, audited by Ford, with source code and verification scripts published in the erdos-frontier-atlas GitHub repo.
- The author cautions the gain is tiny and only holds for enormous k; review so far is AI-only and not refereed, so expert eyes are welcome.
- The repo also includes math audits, constant checks, and symbolic verification materials.
Related event: GPT Astra Helps Advance Lower Bound on Erdős Problem #357(3 posts)→
More from Models
- Qwen3.5 "weight editing" hype debunked: it's just a steering vector plus trial and error — akbirthko · 2026-11-03
- MIT CSAIL shows GPT-6 editing another model's weights by hand, boosting math accuracy with a 0.016% edit — kevrussell · 2026-11-03
- Reproduced GPT patch passes sycophancy test 3/3, but MATH-500 still drops 93% to 33% — kevrussell · 2026-11-03
- Audit update: MATH-500 truncation errors drop from 93% to 33%, but looping persists — kevrussell · 2026-11-03
- Audit of 64,222 labelled answers confirms claimed eval gains: MATH-500 7.9% to 48.1% — kevrussell · 2026-11-03
- Apple reportedly ships a secret multimodal node-graph editor with every iOS copy — round · 2026-11-03