OpenAI researcher Sébastien Bubeck reportedly prompted GPT-5 Pro with a convex-optimization problem, and the model produced a different proof tightening a step-size bound from 1/L to 1.5/L; a later human result reportedly improved the bound further. The supplied snippets characterize the proof as novel and correct, but mostly repeat Bubeck’s claim and do not provide the proof or a clearly identified independent mathematical review. They also refer to GPT-5 Pro—not GPT-5.6—and do not substantiate the claimed 30-year age of the gap, so those parts of the hypothesis remain unsupported here.
If independently validated, this would materially support Scott’s claim that loosely directed human–AI discovery can produce a genuinely new, checkable artifact, especially his Human Pivot + AI Proof and discovery-workshop frameworks. For now, the missing independent proof review, prior-art check, and model-name mismatch make it a promising validation case rather than established evidence.
ip:framework.the-reshape-human-pivot-ai-proofip:framework.discovery-workshop-vs-pr-factoryip:concept.mechanically-different-verifiersip:concept.prior-art-sweep
queries asked of Scott's wikis
- independent verification of AI-generated mathematical proofs
- LLMs as autonomous scientific discovery systems
- reasoning benchmarks versus genuinely novel research
- provenance and novelty checks for model-generated knowledge
- human-in-the-loop workflows for AI research agents
- prompted discovery and artifact-based evaluation of frontier models
2026-07-25T09:26:52Z
No activity since last look; still no independent proof review, prior-art analysis, or resolved model attribution. Nothing left to track within horizon — closing until genuine new evidence (proof artifact or mathematician review) surfaces.
2026-07-22T08:26:56Z
The evidence set still contains only repeated reporting and general reaction, not an independent proof review, prior-art analysis, or inspectable artifact. With the model identity and claimed novelty still unresolved, further discussion activity adds no substantive weight.
2026-07-22T07:25:23Z
The latest activity is repetitive amplification, not mathematical scrutiny; no independent review, prior-art analysis, proof artifact, or clearer attribution has emerged. The claim remains uncorroborated and the modest discussion growth does not warrant renewed attention.
2026-07-20T06:32:16Z
The attached item repeats the same underlying report rather than adding an independent proof review, prior-art analysis, or evidence of the model’s precise contribution. The model-name mismatch and core claims of novelty, correctness, and a 30-year gap remain unresolved.
2026-07-20T06:04:40Z
evidence attached: hn.story.48939768 — This is direct additional reporting on the claimed GPT-5.6 convex-optimization breakthrough.
2026-07-20T04:38:41Z
grounded: converges/medium — If independently validated, this would materially support Scott’s claim that loosely directed human–AI discovery can produce a genuinely new, checkable artifact
2026-07-20T02:21:04Z
The supposed new evidence is only a reobservation of the same unchanged discussion, whose leading comments remain broad AI speculation rather than mathematical review. The proof’s novelty, correctness, and GPT-5.6’s contribution are still entirely uncorroborated.
2026-07-20T01:37:48Z
The refreshed discussion is largely speculative amplification rather than independent mathematical scrutiny. No new evidence establishes novelty, correctness, or GPT-5.6’s substantive contribution, so the case remains an uncorroborated claim.
2026-07-19T11:27:17Z
case created — The extraordinary research claim and intense technical discussion warrant rapid scrutiny for novelty, correctness, and the model's actual contribution.