OpenAI says its models produced a proof challenging a core computing assumption

Key Points

1. OpenAI released new model-generated results in mathematics, computer science and physics, published for outside researchers to examine.
2. One manuscript argues the Exponential Time Hypothesis, a foundation of hundreds of theoretical computer science papers, is false.
3. If the argument holds, assumptions behind algorithms in databases, computational biology and engineering could be revisited.

The latest:

A manuscript posted by OpenAI to GitHub presents a proof claiming the Exponential Time Hypothesis is false, one of two new documents the company released alongside fresh results in mathematics, computer science and physics. The release follows manuscripts running past 700 pages that covered solutions and progress on 372 major mathematical problems, according to a report by Engadget.

Details:

• The two manuscripts: OpenAI added two scientific manuscripts to GitHub and made them available to researchers. The first concerns a proof of Sogge’s Local Smoothing principle, tied to wave equations and specifically the behavior of their solutions across all spatial dimensions, according to the Engadget report reviewed by Al Arabiya Business.
• The contested hypothesis: The second manuscript argues the Exponential Time Hypothesis is wrong. ETH is a central assumption in theoretical computer science, holding that certain complex computational problems cannot be solved by fast algorithms within specific time bounds. It is widely accepted among researchers.
• The stakes: ETH has underpinned hundreds of scientific papers across databases, computational biology and engineering. If it is shown to be false, the result could open the way to more efficient algorithms for a range of complex problems and unsettle assumptions researchers have relied on for years.
• A split approach: The company handled the three fields differently this time. It published the mathematical discoveries directly so researchers could study and interpret them independently, while bringing in outside expert groups to examine and explain the computer science and physics results.
• Computer science: For that field, OpenAI asked its expert group to prepare papers and scientific materials explaining what the models produced, while making the AI outputs themselves publicly available.
• Physics withheld: OpenAI delayed publication of its physics materials until the group assigned to examine them finishes its work. The company did not set a date for that release.
• Who is reviewing: The scientists who originally proposed the Exponential Time Hypothesis are themselves part of the expert group OpenAI brought in to examine and explain the output of its AI models, placing the authors of the assumption in the position of assessing a claim against it.
• Open invitation: OpenAI said it welcomes feedback on how to widen participation in the process and make it more useful to the scientific community, without naming which institutions or individuals it expects to take part.

Background:

The release builds on an earlier batch of OpenAI manuscripts exceeding 700 pages that reported solutions and partial progress on 372 significant mathematical problems, a volume that pushed the question of independent verification to the front for outside researchers.

Between the lines:

The differing treatment of the three fields reads as a calibration of confidence: mathematics published raw for independent scrutiny, computer science routed through explainers, physics held back entirely. Handing the ETH claim to the hypothesis’s own authors sets the highest possible bar for a result that, if validated, would disturb hundreds of papers built on it.

What’s next

Watch for the expert group’s assessment of the ETH proof, the delayed release of the physics materials, and whether independent researchers reproduce or refute the Sogge’s Local Smoothing argument on GitHub.