Vibeleaderboard
← All Intel
Intel / post

Breakthrough Results in Sphere Packing, Coding Theory, and Group Theory

Source
x.com
Date
OpenAI@OpenAI
Thread · 2 parts

An internal version of our next major model produced 10 new results on long-standing open problems in mathematics and theoretical computer science, using roughly $2,000 worth of tokens at GPT-5.6 Sol API rates.

The results span sphere packing, coding theory, group theory, quantum complexity, lattice cryptography, extremal combinatorics, and more. Among them: establishing the existence of non-sofic groups and exponential improvements to bounds on high dimensional sphere packing.

Context

OpenAI says an unreleased internal version of its next major model produced 10 new results on long-standing open problems in mathematics and theoretical computer science. The evidence available here supports that only as OpenAI's claim: the announcement thread names the fields and two of the results, but the proofs and any technical write-up were not part of the material read.

The thread lists results in sphere packing, coding theory, group theory, quantum complexity, lattice cryptography and extremal combinatorics, and highlights two. One establishes that non-sofic groups exist. Sofic groups are a broad class of groups that can be approximated by finite permutations, and whether every group belongs to that class has been an open question, so an existence proof would answer it in the negative. The other is an exponential improvement to bounds on sphere packing in high dimensions, the problem of how densely identical balls can fill space. The thread does not say which bound improved, whether the density known to be achievable or the ceiling on what is possible.

OpenAI puts the cost at roughly $2,000 worth of priced at GPT-5.6 Sol's API rates, not at any price for the internal model itself. That figure describes model usage only. The thread does not say how many attempts were made, how much people guided the work, or whether outside mathematicians have checked the results. Collection was also incomplete: the thread read was cut short by a reading budget, and the attached video yielded no text. Until the proofs can be read and checked by others, these are announced results, not established theorems.

Terms in this piece · Glossary
  • token — The chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.
More from OpenAI
Recommended reads
Comments

Checking sign-in…

Loading comments…