Issues

• MAIS-O28: intended work on the local zero-regret exception-mass modulus

  Math-for-AI-Safety/MAIS#45 · Filter by author mimosapudical opened on Oct 17, 2026

• MAIS-O54: Refutation of the unconditional conjecture

  A5correctionprogressunscreened
  Math-for-AI-Safety/MAIS#44 · Filter by author berczig opened on Oct 4, 2026

• MAIS-O53: Refutation of Conjecture 5.3 on ratio of learned pure irreps in group multiplication NNs

  A5candidate-solutionunscreened
  Math-for-AI-Safety/MAIS#43 · Filter by author berczig opened on Oct 4, 2026

• MAIS-O68: Learning coefficient and multiplicity for single attention layer (using AlphaEvolve, Astra Ultra and Fable 5.1)

  A6candidate-solutionunscreened
  Math-for-AI-Safety/MAIS#42 · Filter by author berczig opened on Oct 4, 2026
  1

• MAIS-05 -- is the distribution well-defined?

  A5correctionunscreened
  Math-for-AI-Safety/MAIS#41 · Filter by author geordw opened on Oct 2, 2026

• Applying AI-based mathematical problem solving to XAI and AI alignment

  discussionmetaunscreened
  Math-for-AI-Safety/MAIS#40 · Filter by author amugea opened on Sep 28, 2026

• New reference summary: [L26] Larsson, Langevin dynamics along the zero set of real-analytic potentials

  metareferenceunscreened
  Math-for-AI-Safety/MAIS#39 · Filter by author martin-o-larsson opened on Sep 23, 2026

• MAIS-O58: corrections to the exchangeability formulation, with an elementary p = 7 case

  A5candidate-solutioncorrectionunscreened
  Math-for-AI-Safety/MAIS#38 · Filter by author ChengXinbu opened on Sep 19, 2026

• MAIS-O47, part (2): an elementary candidate solution via strong regularization

  A4candidate-solutionunscreened
  Math-for-AI-Safety/MAIS#37 · Filter by author ChengXinbu opened on Sep 19, 2026

• MAIS-O36: candidate negative resolution — the conjecture is false as stated, at usage probabilities very small relative to the penalty

  A3ai-generatedcandidate-solutionunscreened
  Math-for-AI-Safety/MAIS#36 · Filter by author 26david26 opened on Sep 17, 2026

• MAIS-O87: candidate negative resolution — the probe logit is eventually negative at every positive diversity

  A8candidate-solutionunscreened
  Math-for-AI-Safety/MAIS#35 · Filter by author Robby955 opened on Sep 13, 2026
  1

• MAIS-O43: partial measurements, a dominant diffuse phase (no recovery, no merging)

  A3progressunscreened
  Math-for-AI-Safety/MAIS#34 · Filter by author alexisdpc opened on Sep 11, 2026

• MAIS-O40: candidate complete solution — every global minimizer is a regular pentagon for all sufficiently small activation probability

  A3ai-generatedcandidate-solutionunscreened
  Math-for-AI-Safety/MAIS#33 · Filter by author 26david26 opened on Sep 8, 2026

• Two open problems from Menon, "The geometry of the deep linear network" (arXiv:2411.09004), for open-problems/

  metanew-problemunscreened
  Math-for-AI-Safety/MAIS#32 · Filter by author garry-gum opened on Aug 31, 2026

• MAIS-O24 has no solution, so Conjecture 4.4 (MAIS-O26) has no instances — is "𝒩 as in Problem 4.3" the intended reading?

  A2correctionunscreened
  Math-for-AI-Safety/MAIS#31 · Filter by author mbrcic opened on Aug 30, 2026

• MAIS-O38 candidate complete solution: m^3+2m fixed codes suffice for every sparsity k < m

  A3ai-generatedcandidate-solutionunscreened
  Math-for-AI-Safety/MAIS#30 · Filter by author 26david26 opened on Aug 26, 2026
  1

• MAIS-O66 candidate complete solution: generic constancy holds, with an explicit polynomial h

  A6ai-generatedcandidate-solutionunscreened
  Math-for-AI-Safety/MAIS#29 · Filter by author 26david26 opened on Aug 26, 2026

• MAIS-O86 candidate complete solution: the limit is 1, not the Gaussian process posterior Q(L)

  A8ai-generatedcandidate-solutionunscreened
  Math-for-AI-Safety/MAIS#28 · Filter by author 26david26 opened on Aug 25, 2026

• MAIS-O59(ii): the exact-purity clause is false — a certified mixed-frequency local margin maximiser

  A5ai-generatedcandidate-solutionunscreened
  Math-for-AI-Safety/MAIS#27 · Filter by author 26david26 opened on Aug 24, 2026

• MAIS-O82: candidate complete solution — the proxy wins at zero diversity, at every step size

  A8ai-generatedcandidate-solutionunscreened
  Math-for-AI-Safety/MAIS#26 · Filter by author 26david26 opened on Aug 24, 2026

• MAIS-O83: candidate negative resolution — no coin-lover in the standard family

  A8ai-generatedcandidate-solutionunscreened
  Math-for-AI-Safety/MAIS#25 · Filter by author 26david26 opened on Aug 24, 2026

• MAIS-O29(a): machine-checked negative answer — boltzmann behavior is not injective

  A2candidate-solutionunscreened
  Math-for-AI-Safety/MAIS#24 · Filter by author mbrcic opened on Aug 24, 2026

• MAIS-O16 / MAIS-A1 Problem 4.8: preliminary empirical code

  A1help wantedprogressquestionunscreened
  Math-for-AI-Safety/MAIS#23 · Filter by author jackringel opened on Aug 21, 2026

• MAIS-O63 Part a

  A6progressunscreened
  Math-for-AI-Safety/MAIS#22 · Filter by author iris-garvy opened on Aug 19, 2026

• MAIS-O6: Numerical counterexample

  A6progressunscreened
  Math-for-AI-Safety/MAIS#21 · Filter by author iris-garvy opened on Aug 14, 2026