Peres’s weighted-majority noise question (Codex, Claude)
Awaiting review
The takeaway
On 21 November 2026 Yitzchak Shmalo posted on Zenodo a proof that every linear threshold function changes value with probability at most (2/π)√ε + Cε^(3/5) under ε-noise.
Status
- Claim
Awaiting review
- Our reporting
- Low confidence
- Verification
- Preprint only
- Importance
- 3 of 5
- Last verified
- 21 November 2026
Your AI and this story
- GPT-6.2 Astra174 days after its cutoff
- Claude Opus 5.5144 days after its cutoff
- Gemini 4 Hydrogen235 days after its cutoff
- Grok 4.7174 days after its cutoff
None of these four assistants can know about it. The closest, Claude Opus 5.5, stops 144 days before it.
Key facts
- Claim (Zenodo 24414132): NS_ε(f) ≤ (2/π)√ε + Cε^(3/5) for every linear threshold function f on {−1,1}^n, with an absolute constant C; also characterises the sequences that attain the constant
- Background: Peres proved the bound 2√ε (asymptotic constant √(2/π)) and asked whether 2/π, the value for simple majority, is the right constant; the Majority Is Least Stable conjecture is due to Benjamini, Kalai and Schramm
- Tools named in the proof: the Berry–Esseen theorem, Erdős’s Littlewood–Offord bound and Servedio’s critical index
- AI declaration: ‘This paper is part of a broader research project on proving open problems with AI systems. The results, proofs and text were produced with generative AI systems. OpenAI’s Codex wrote the proofs, and Anthropic’s Claude (Opus 5.5) checked them and wrote the text’; the agents ran inside Shmalo’s ‘Math Brain’ framework
- Code and supplements: github.com/yspennstate/weighted-majority-noise-constant
What happened
Yitzchak Shmalo’s Zenodo deposit (21 November 2026) is part of his project of attacking open problems with networks of AI agents (“Math Brain”). The paper credits Codex with the proofs and Claude Opus 5.5 with checking them and writing the text. It reached us through the sweep’s Zenodo search for “conjecture” together with model names.
Why it matters
Peres’s question is a well-known open problem in the analysis of Boolean functions and appears on O’Donnell’s list as a weaker form of the Majority Is Least Stable conjecture. The claim has no formal proof and no expert review yet.
What is disputed or not yet verified
| Verification | Unverified preprint on Zenodo (no Lean) |
|---|
Sources
2 sources from 2 sites. Numbers match the chips in the text.
2 sources: 2 primary
Primary
- Zenodo 24414132: The sharp small-noise constant for weighted majority (Shmalo)zenodo.org, paper
- GitHub yspennstate/weighted-majority-noise-constantgithub.com, code
Changes
- Filed (science agent; Zenodo record and PDF)