anthropic.com via Hacker News

Anthropic Says Claude Solved Two Open Black Hole Problems

TL;DR

  • Anthropic says a research group of Claude agents, run in the background for about six weeks, produced solutions to two open theoretical-physics problems tied to black holes and quantum gravity.
  • One result completes a calculation set out in 2003 on the Hawking-Page phase transition; the other proves the lightest 3D-quantum-gravity black hole can weigh at most c/Λ, with Λ ≈ 9.1266.
  • A physicist checked every main result and the proofs were formally verified in Lean in September; Anthropic says neither result describes our universe or has practical applications.

Anthropic says its Claude model has solved two open problems in theoretical physics, both tied to black holes and the search for a theory of quantum gravity, according to a research post the company published on anthropic.com. "Claude has solved two open problems in theoretical physics," the post opens. "Both concern black holes and the search for a theory of quantum gravity."

The first problem traces to a 1983 paper by Stephen Hawking and Don Page on a phase transition in how black holes form, which the post likens to "how water suddenly boils at 100°C." Anthropic writes that "in 2003, a group of physicists set out the calculation that would settle the question, but it was extremely complex and was never completed." Claude finished it.

The second concerns the modular bootstrap, a technique physicists have used since 2009 to bound the lightest black hole in a 3D model of quantum gravity. "Claude found the spectral curve that governs the modular bootstrap at large values of c," the post states, and used it to prove the lightest such black hole "can be no heavier than c/Λ, where Λ is given by an exact formula and is approximately 9.1266." An earlier Claude had failed on the same problem across 61 parallel attempts.

The setup was informal. "In August, an Anthropic researcher, Hongbin Chen, was testing Claude's capabilities as a side project," the post says. "The agents ran in the background, with no special setup, for about six weeks." Chen's agents worked as copies of one another: "Claude ran a small research group of agents, all copies of itself. Almost every piece of the calculation was done twice, by agents acting independently." Afterward, "Claude then wrote the paper with minimal guidance, and a physicist checked every main result," and "in September, the results were formally verified in Lean, a computerized proof checker." The physicist is not named in the post.

Anthropic is careful about scope. "Neither result describes our own universe directly, and neither has practical applications, but both contribute to a long-running theoretical effort," the post reads, closing with: "Neither result settles the question of quantum gravity, or even comes close to doing so, but each pushes physics forward in a longstanding area of research." The release lands in a busy stretch of company news this week, the same day Anthropic shares reportedly moved after-hours on reports of Apple's own AI agent plans.

Shared on Bluesky by 1 AI expert

  • Stefanie Hane @stefihane.bsky.social amplified

    @ai-nerd.bsky.social

    wait, an earlier Claude agent failed in 61 parallel attempts on this black hole bound, and half of the proof that finally worked borrows the strategy of a sphere-packing proof by an OpenAI model https://www.anthropic.com…

    View on Bluesky →