In 1917 Sōichi Kakeya asked a question you can try with a pencil: what is the smallest region in which a needle can be turned all the way round? In 1928 Abram Besicovitch showed the answer is as small as you like. The real question became what such sets must still look like, and the Kakeya conjecture says that a set containing a line segment in every direction must be, in a precise sense, as large in dimension as the space around it. It was proved in the plane in 1971. In February 2025 Hong Wang and Joshua Zahl proved it in three dimensions, and in July 2026 Wang received a Fields Medal; the field described higher dimensions as open, the obvious next target. On October 26, OpenAI released a 59-page paper in which its model claims Sogge’s 1991 local smoothing conjecture in every dimension, and with it the Kakeya conjecture, the Bochner–Riesz conjecture and other long-standing problems, with Lean statements for six theorems. It released the model’s output directly, leaving the explanation to researchers. No specialist has yet given a verdict.
Sōichi Kakeya posed the needle problem in 1917. A unit needle can be turned inside a disk of area π/4, or inside a three-pointed deltoid of half that; in 1920 Gyula Pál showed that among convex regions the best is an equilateral triangle. Then Abram Besicovitch showed that if the region need not be convex, the area can be made as small as you like, and that sets containing a segment in every direction can have zero area.[6] The companion page lets a reader try both.[10]
Zero area does not mean small in every sense. The Kakeya conjecture says that a set containing a unit segment in every direction must have full dimension: in the plane, dimension 2; in space, dimension 3. Roy Davies proved the planar case in 1971.[6] Three dimensions stood for decades until February 2025, when Hong Wang and Joshua Zahl posted a proof.[4] In July 2026 Wang received a Fields Medal. Quanta’s report that day described the community’s mood — “the holy grail has been achieved in some sense” — and the next step as open: the four-dimensional case was the obvious target, and “no one knows what that would look like yet.”[5]
OpenAI’s October 6 release already included model-written manuscripts on the Kakeya maximal conjecture in three dimensions and on Kakeya sets in four.[2] On October 26 it went further. Its paper, “Local smoothing for the Euclidean wave equation,” dated October 25, claims Sogge’s local smoothing conjecture of 1991 in every dimension at least two. From it, the paper deduces the Kakeya and Nikodym maximal conjectures, so that “every Euclidean Kakeya set has full Hausdorff dimension,” as well as the Bochner–Riesz conjecture and Fourier restriction estimates throughout the interior of the conjectured range.[1] Before this, local smoothing was known at the critical exponent only in the plane, by Guth, Wang and Zhang in 2019.[9]
The formal side is substantial. OpenAI’s Lean comparator file states six theorems — local smoothing, Bochner–Riesz, spherical restriction, the Kakeya maximal bound, the Kakeya set conjecture and Stein’s square function — and the release checks proofs of them against those statements.[3] The release note adds how the paper was published: “We are releasing the model output directly, as we did in our October 6 release.”[2] Coverage of the release describes the write-up and the Lean proof as public, alongside earlier results in two and three dimensions.[7]
The three results of October 26 were released three ways. The computer-science proof went to a working group led by the hypothesis’s authors; the physics results were withheld for the physicists to write up; the mathematics was published as the model wrote it.[2] For the Kakeya paper, the explanation is deliberately left to people. A Lean check can confirm that a stated theorem follows from its premises. It cannot confirm that a formal statement says what the field means by the Kakeya conjecture, and it cannot make a 59-page argument understood.
That is the gap this case watches. In UC-339 the hypothesis’s authors are checking the claim. Here, the people best placed to read the claim are those who did the human work: Wang, Zahl, Guth and the harmonic analysts who spent decades on these estimates. Their readings, not the release, will decide how the field receives it. As of this writing, no public verdict has appeared.
The honest limits. The paper is an unreviewed preprint written by a model. The claim is far-reaching — local smoothing, Kakeya, Bochner–Riesz and restriction at once — which makes it either one of the larger advances in harmonic analysis in decades or a long argument with an error somewhere that no formal check has caught. Only careful human reading can tell which, and that takes time.
Sets with a segment in every direction can have zero area; needle sets can be arbitrarily small.[6]
SurpriseNo one knows what that would look like yet. — Quanta Magazine, on the four-dimensional Kakeya conjecture, July 23, 2026 [5]
| Dimension | Evidence |
|---|---|
| Quality (D5) Origin · 86 | Local smoothing for n ≥ 2, implying Kakeya, Nikodym, Bochner–Riesz and restriction in the interior range.[1]Every Dimension at Once |
| Employee (D2) L1 · 80 | Three months after Wang’s Fields Medal, higher dimensions were “the obvious next target.”[5]The Field’s Next Target |
| Operational (D6) L1 · 72 | “We are releasing the model output directly”; Lean statements for six theorems.[2][3]Released Raw |
| Customer (D1) L2 · 68 | Restriction and Bochner–Riesz estimates are used across analysis and partial differential equations.[1]Tools for Many Fields |
| Regulatory (D4) L2 · 66 | No specialist verdict yet; the model output was released directly, without a human explanation.[2]Accepting the Unexplained |
The cascade originates in D5 (Quality): a claimed proof of local smoothing in every dimension, with Kakeya, Bochner–Riesz and restriction as consequences.[1] It propagates to D2 (Employee), the mathematicians whose next target this was, three months after a Fields Medal for the three-dimensional case,[5] and to D6 (Operational), how the result was released — raw, with Lean statements for six theorems, explanation left to people.[2][3] It lands in D4 (Regulatory), the norms by which a field accepts a proof nobody has explained, and in D1 (Customer), the readers and researchers for whom these estimates are tools. D3 (Revenue) is not scored.
It belongs with UC-339, the computer-science result from the same release, and with UC-328 and UC-334 from the October 6 release, where specialists read and extended machine-made proofs within days. The pattern holds: answers arrive first, and understanding follows at the speed of people. The review date in January will look for the first specialist readings.
-- UC-340: Every Dimension, Three Months Later: 6D Amplifying Cascade
-- Kakeya 1917; Pal 1920 convex minimum equilateral triangle; Besicovitch 1920 and 1928 arbitrarily small and measure-zero sets; Davies 1971 plane. Sogge local smoothing 1991. Guth, Wang, Zhang 2019 plane. Wang and Zahl Feb 2025, arXiv 2502.17655, Kakeya set conjecture in three dimensions; Hong Wang Fields Medal Jul 23 2026; Quanta: holy grail achieved in some sense, four dimensions the obvious next target, no one knows what that would look like yet. OpenAI Oct 6 release: Kakeya maximal 3D and four-dimensional Kakeya sets. OpenAI, Local smoothing for the Euclidean wave equation, dated Oct 25, released Oct 26 2026, 59 pages: local smoothing for n at least 2; deduces Bochner-Riesz, Stein square function, restriction in the interior range, Kakeya and Nikodym maximal; every Euclidean Kakeya set has full Hausdorff dimension. Lean comparator states six theorems. Released model output directly, explanation left to researchers. No specialist verdict yet.
FORAGE every_dimension_three_months_later
WHERE central_conjecture_claimed_all_dimensions = true
AND human_proof_lower_dimension_recent = true
AND output_released_raw = true
ACROSS D5, D2, D6, D1, D4
DEPTH 3
SURFACE every_dimension_three_months_later
DIVE INTO claimed_not_explained
WHEN specialist_verdict_public = false
AND lean_statements_public = true
TRACE proof_acceptance_cascade
EMIT every_dimension_three_months_later_signal
DRIFT every_dimension_three_months_later
METHODOLOGY 88
PERFORMANCE 40
FETCH every_dimension_three_months_later
THRESHOLD 1000
ON WATCH CHIRP high 'Three months after Hong Wang's Fields Medal for the three-dimensional Kakeya conjecture, an OpenAI model claims local smoothing and Kakeya in every dimension, with Lean statements for six theorems, released as raw model output for mathematicians to explain. No specialist verdict yet.'
SURFACE analysis AS jsonRuntime: @stratiqx/cal-runtime · Spec: cal.semanticintent.dev · DOI: 10.5281/zenodo.18905193
A set can turn a needle in almost no area and still be as large as the space in dimension. That is the conjecture, and it is subtle.[6]
The field called higher dimensions open in July. In October a machine claims all of them at once.[5][1]
The output was published raw, as the model wrote it. Explaining it is now the mathematicians’ work, and it cannot be rushed.[2]
Lean confirms six formal statements. Whether they say what the field means, and why they are true, is a human question.[3]
Primary sources are OpenAI’s paper and its Lean comparator statements, read at source, OpenAI’s release notes, and Wang and Zahl’s 2025 preprint. The paper is unreviewed. Source 10 is an interactive companion made for this case.
This case follows a far-reaching proof released as raw output, three months after the field celebrated a smaller human step. Organisations face the same choice with machine-made answers: use them, check them, or wait. Book a discovery call to decide which answers you can act on.