Yes, Claude can do nine loops — SkimNews
Get the Health newsletter
Daily health & science — research, biotech, public health, the studies worth knowing. Free.
- Anthropic physicists Liam Fitzpatrick and Siddharth Mishra-Sharma used Claude via the Claude Science platform and Fable 5.1 to compute the six-particle hexagon amplitude in planar N=4 super Yang-Mills at nine loops, reaching out to von Hippel at the end of August.
- Claude tackled the calculation two independent ways — the original bootstrap method and an indirect form-factor approach — each costing an end-user roughly $1,000–$2,000, with the bootstrap portion consuming only about $100 of compute (96 CPUs running for one week).
- The bootstrap technique resembles solving a Sudoku puzzle: researchers specify constraints the answer must satisfy and let the system narrow possibilities — a structure von Hippel called "bizarrely well-suited" for LLM-driven workflows.
- Matt von Hippel, a former theoretical physicist turned science writer, issued the challenge on his blog after wanting to see an LLM take on a problem that was computationally hard on an academic budget, not just one limited by missing ideas.
- Song He's group at the Chinese Academy of Sciences independently reached most of the same result around the same time, using some GPT-6 AI assistance but with a more human-in-the-loop approach than Anthropic's near-autonomous run.
- Lance Dixon, the amplitudeologist who found the eight-loop result and expected nine loops would require a different indirect method, verified Anthropic's answer.
Why it matters: Anthropic showed a frontier scattering amplitude problem — one amplitudeologists expected to need new indirect techniques — could be solved by an LLM on an academic-scale budget (~$1,000–$2,000, 96 CPUs for a week). A competing Chinese group reached the same answer within days using GPT-6, indicating AI-assisted bootstrapping is becoming a repeatable research workflow in the field rather than a one-off stunt.
Ask SkimNews
