← Wider interest

From a conversation to two lines of enquiry

Three, seven & the future
of mathematics.

An ordinary operation with an extraordinary exception—and a mathematician forecasting how AI could change the way we verify what we know.

Written by GPT 6 Astra, at Lawrence Rowland’s request.

13 November 2026 · Based on the conversation and sources consulted 10 November 2026

The conversation

Curt Jaimungal’s Theories of Everything brought together Eric Weinstein and Fields Medalist Richard Borcherds in “Fields Medalist & Eric Weinstein on the Future of Mathematics”, publicly released on 9 November 2026. Two details invite a closer look: the cross product’s unusual dimensional restrictions, and Weinstein’s endorsement of Stanford mathematician Jared Duker Lichtman.[1]

Two possible dimensions.
Three and seven.

Weinstein’s wording can catch the ear: the vector cross product exists in “two dimensions”. He means two permissible dimensionalities, namely 3 and 7—not the two-dimensional plane.[1]

The familiar cross product takes two arrows, u and v, and returns another arrow in the same space. Three requirements do the work:

  • The result is perpendicular to both inputs.
  • Its length equals their parallelogram’s area.
  • It is bilinear: addition and scaling in either input pass consistently through to the result.
‖u × v‖ = ‖u‖ ‖v‖ sin θθ is the angle between the vectors, from 0° to 180°.

Swapping the inputs reverses the result: v × u = −(u × v). These requirements together are far more restrictive than simply finding a perpendicular arrow.[2]

Try the 3D case
uvproduct
Two vectors, their parallelogram and a perpendicular cross productA perspective projection of unit vectors u and v in the xy plane. At 60 degrees, the parallelogram has area 0.866, and u cross v points up the z axis with length 0.866.zxyθuvu × v0
Parallelogram area0.866
Cross-product length0.866

Same area and length; reversing the order reverses the arrow.

A perspective view of 3D space, not a 7D illustration. Here u = (1, 0, 0) and v = (cos θ, sin θ, 0). The orange arrow points perpendicular to their plane; the numbers show true lengths and area, not projected screen measurements.
2

No room for the result

Two independent vectors span the whole plane. No nonzero vector within that plane is perpendicular to both. A signed-area scalar is possible, but is a different output.

3

One perpendicular line

The vectors’ plane leaves a single perpendicular line. The right-hand rule chooses a sense; the area fixes the length.

7

An extra structure chooses

There are many perpendicular directions. Octonionic multiplication selects a result consistently with the algebraic requirements.[2]

Seven is eight, minus the real part.

Hurwitz’s theorem supplies the deeper restriction. Finite-dimensional real normed division algebras, with a unit and bilinear multiplication satisfying ‖ab‖ = ‖a‖ ‖b‖, occur only in dimensions 1, 2, 4 and 8.

Separate out the one real direction. The remaining “imaginary” part has one fewer dimension and inherits a cross product. For quaternions and octonions, it is the imaginary part of the product of two imaginary elements.

Crucially, the reasoning runs both ways. A cross product obeying the stated conditions lets us add one real dimension and construct a normed division algebra. Hurwitz therefore excludes the missing dimensions; this is not merely a list of examples found so far.[3]

The algebra in one line

For real scalars a, b and vectors u, v, define a multiplication on ℝ ⊕ V:

(a, u)(b, v) = (ab − u · v, av + bu + u × v).

The cross product’s identities make the norm multiplicative. Conversely, recover u × v from the imaginary part of uv. The bridge is a construction in both directions.[3]

The four possibilities
Number systemTotal
dimensions
Imaginary
dimensions
Real numbers · ℝ10
Complex numbers · ℂ21
Quaternions · ℍ43
Octonions · 𝕆87

The 0D and 1D cross products are identically zero. The nontrivial cases are 3D and 7D.[4]

Jared Duker
Lichtman

Szegő Assistant Professor of Mathematics
Stanford University

His research concerns multiplicative number theory: primes and the subtle structures arising from multiplication. He completed his Oxford doctorate under James Maynard in 2023.[5]

His AI involvement includes work with Gauss, an agent used to obtain a Lean formalisation of the strong Prime Number Theorem. He presented that work at the Fields Institute in October 2025.[6]

A published forecast · September 2026

A machine-checkable library
of known mathematics.

In his Mathematics Autoformalization Project essay, Lichtman argues that translating the known mathematical literature into formal code could become achievable within a year, given sufficient coordination and funding.[7]

Autoformalisation means using AI to translate mathematics into a precise language whose proofs can be checked by a proof assistant such as Lean. Formalising an existing proof and discovering a new theorem are different achievements.

This is Lichtman’s dated forecast and proposed programme, not an established completion timetable.

At 1:48:16 in the conversation, Weinstein says Lichtman has repeatedly been right on matters where Weinstein thought him wrong. The endorsement supplies no forecast-by-forecast record. It is a reason to examine Lichtman’s work, rather than independent proof of his predictive accuracy.[1]

Sources & further reading

Links distinguish the conversation, the mathematics and Lichtman’s own claims. The forecast is preserved in its historical context.

  1. The conversation and transcriptTheories of Everything, Richard Borcherds and Eric Weinstein. Public release 9 November 2026; transcript hosted by The Portal Wiki. Lichtman reference: 1:48:16.
  2. John C. Baez · The Octonions, §4.1Cross products, imaginary quaternions and octonions, and the role of G₂. From the 2002 Bulletin of the American Mathematical Society survey.
  3. Bruce W. Westbury · Hurwitz’ Theorem2009. Especially §2: composition algebras and vector product algebras; the construction adjoining a real unit.
  4. Pavlek & Perše · Cross product on ℝⁿ, normed algebras and H-spaces2016. The classification 0, 1, 3, 7 and its relationship with the normed-algebra dimensions 1, 2, 4, 8.
  5. Stanford · Jared Duker LichtmanInstitutional biography and academic appointment; research interests on his linked personal homepage.
  6. Gauss: an agentic formalization of the Prime Number TheoremFields Institute talk, 28 October 2025. Abstract describes the Lean proof and the challenge set by Alex Kontorovich and Terence Tao.
  7. Lichtman · The MAP: Mathematics Autoformalization ProjectSeptember 2026 essay, read via a mirror of his X article. Original post. A primary statement of his ambitions and forecast.