Which of OpenAI's claimed math results will be shown to be flawed before 2027?

MANIFOLD

Ṁ2.3kṀ3.6k

Dec 31

Quasi-Riemann Hypothesis: no zeros of ζ or any Dirichlet L-function for Re(s) > 7/8

39%

Log abundance + Iitaka subadditivity + minimal models (034/033/036)

34%

Rational Hodge conjecture for CM abelian varieties (032)

ω ≤ 9/4 over ℂ for matrix multiplication (107)

15%

FFT in O(n(log n)^(1−10⁻¹³)) (130)

22%

Integer multiplication in O(n (log n)^(1−2⁻¹⁸²)) (109)

Unique Games Conjecture (102)

Critical percolation on ℤ³ (213)

37%

3D Fourier restriction + Bochner–Riesz (077/078)

34%

Anderson localization/delocalization (261)

33%

Hilbert's 16th problem, uniform bound on limit cycles (143)

Irrationality exponent of π is 2 (017)

28%

BEC at positive temperature (267)

29%

Stationary varifold singular sets (346)

30%

4D disk embedding fails (305)

28%

Nearby Lagrangian conjecture disproved (340)

28%

Hilbert–Smith (304)

Rokhlin multiple mixing (145)

39%

Hilbert's 10th problem over ℚ is undecidable (004)

15%

Kaplansky counterexamples + non-sofic group (196/197)

Resolves YES if a flaw is found, even if OpenAI fixes it. Bundled answers resolve YES if any one component fails. A gap between a Lean statement and its paper does not count as a flaw, the flaw must be in the correctness of the proof. Timezone AoE if that matters. Flaw claim should be acknowledged by OpenAI, or stand unrebutted for about two weeks.

I will not bet after setting probabilities.

The creator has blocked themselves from betting in this market.

Technical AI Timelines

OpenAI

AI Impacts

AI Regulation

Get

1,000

to start trading!