Urgent.News

What's breaking now, across thousands of outlets.

AI

The most exciting claims from OpenAI’s heap of new proofs

Scientists have called out several math and computer science results as the most significant in the company’s overwhelming new deluge of achievements

OpenAI's recent release of over 700 manuscripts containing proofs, solutions, and progress on 372 open problems across various fields of mathematics, theoretical computer science, and physics has generated significant excitement among scientists. While many of these proofs require further verification, experts highlight several key achievements.

One of the most notable results is progress on the Riemann hypothesis, a longstanding unsolved problem in mathematics. The Riemann hypothesis concerns the zeros of the Riemann zeta function, which are intimately connected to the distribution of prime numbers. OpenAI's model has produced two important findings: a proof of the "quasi-Riemann hypothesis," which offers substantial insights into the distribution of prime numbers, and evidence that none of the generalizations of the zeta function possess a "Siegel zero." The latter finding effectively settles the question.

Another groundbreaking result is the resolution of Hilbert's 10th problem. This problem asks whether there exists an algorithm capable of determining whether any given equation has whole-number solutions. Yuri Matiyasevich's work in 1970 proved that such an algorithm does not exist. However, mathematicians have since explored extending the result to equations with a broader range of values.

OpenAI's model has demonstrated that even when variables are allowed to be rational numbers, the set of equations with whole-number solutions remains unsortable.

OpenAI's work also sheds light on the Kakeya conjecture, which involves sliding a piece of chalk around on a surface to cover the smallest possible area while pointing in every direction. In three dimensions, the problem was resolved by Hong Wang and Joshua Zahl this summer. Now, OpenAI's model has extended the solution to four dimensions.

Other important results include proofs related to Artin's conjecture on primitive roots, which involves special number systems with a highest number, and progress on Paul Erdős' famous conjecture, as well as contributions to the Langlands program, a vast set of conjectures often referred to as the "Grand Unified Theory of Mathematics."

Written by urgent.news from Scientific American's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.

Also reported by 1 other outlet

Read the original at scientificamerican.com →

More in AI

The agent finished. Who turns off the VM?

Suppose a coding agent opens a pull request at 6 p.m. The tests pass, a preview is running, and the reviewer has already logged off. The agent's task is complete.

  • Agent session ends when task result is saved
  • Preview retained only until explicit expiry time
  • Workspace removal depends on commit status

Your LLM Types One Token at a Time. It Doesn't Have To.

Every token your LLM emits costs one full forward pass through the entire model. Seventy billion parameters loaded from memory, multiplied, discarded — for a single token. Then again. And again.

  • Speculative decoding drafts tokens with cheap model before big model verification
  • Acceptance rule accepts drafted token with probability min(1, q(d)/p(d))
  • EAGLE-3 achieves 2-3x speedup over vanilla decoding, but diminishing returns at high batch sizes

More from Thursday 8 October →