Can you recognize the integers among the rationals using a polynomial equation? On November 5th Eliot Glazer claimed you can! He said there's a polynomial P in n+1 variables, with rational coefficients, such that q₁ is an integer if and only if there exist rationals q₂, ..., qₙ such that
P(q₁, q₂, ..., qₙ) = 0
The catch: his polynomial had 292,890,231,905 variables! Also, all he could say about its degree is that it's at most 2^2^2^2^2^2^2.
The other catch: Glazer doesn't claim to understand the proof, he got it using AI, the paper is 148 pages long, it hasn't been formalized using Lean, and it would disprove a conjecture by the famous number theorist Barry Mazur.
This might be enough to make you shrug in disgust. But other mathematicians and their AI agents jumped in and started working on this problem. In four days they brought down the number of variables to 962, and the degree of the polynomial to 4.
As of today, they claim to have gotten it down to a degree-4 polynomial with just195 variables.
https://github.com/elliotglazer/diophantine-z-q/releases/tag/
quartic-195-version-20
The catch: its coefficients use the fixed integer C = 149^(BB(2^1000)). Here BB is the Busy Beaver function, and BB(2^1000) is an incomprehensibly large number. Indeed, even BB(745) is beyond our ability to understand using ordinary set theory. (This is a long story.)
The other catch: it still hasn't been verified in Lean.
The other catch: all these results depend on a result in mathematical logic claimed by OpenAI, but not yet verified in Lean. Namely, that no algorithm can decide whether a multivariable polynomial with rational coefficients has a rational zero.