Yes, Claude can do nine loops
Physicist Matt von Hippel challenged Claude to compute a nine-loop scattering amplitude in theoretical physics, a frontier problem it successfully solved using standard bootstrap methods.
summary is enough The post details an impressive autonomous AI physics calculation, but the core narrative and takeaways are fully captured by a summary.
summary
Theoretical physicist Matt von Hippel challenged AI to solve a frontier particle physics problem: computing the nine-loop N=4 super Yang-Mills scattering amplitude. Using Anthropic's Claude Science harness, the model ran autonomously with minimal supervision, successfully generating the solution using standard bootstrap techniques. The result was independently validated by Stanford physicist Lance Dixon, who noted Claude successfully automated a fragile workflow previously thought too difficult to do directly. While not a demonstration of sci-fi superintelligence, it demonstrates that current LLMs can reliably execute complex, multi-step scientific computations.
key points
- Claude solved the nine-loop N=4 super Yang-Mills amplitude problem autonomously via an agentic harness on modest hardware resources.
- Stanford physicist Lance Dixon independently validated the result, noting the model executed a fragile recipe previously thought too difficult for direct automation.
- A competing team at the Chinese Academy of Sciences independently achieved a similar milestone concurrently using GPT-6 assistance.
who should read it
Physicists, researchers, and technical readers interested in practical AI applications for scientific computing.
discussion mixed
consensus
Commenters acknowledge the technical milestone of solving a complex physics problem, but many express heavy skepticism regarding Anthropic's corporate PR framing and marketing angle.
pushback
Some commenters question how much hidden compute, trial-and-error inference, and upstream training costs were required to achieve the claimed one-shot result.
Notwithstanding the duplication of these posts across social media, OpenAI, and Anthropic ("it's all the model, they just tell it to keep going" if you beat anything with RL enough ... it's going to do the thing)
Here's my issue with this post:
> True to the spirit of the challenge, they didn’t use millions of dollars in computer power. They used Fable 5.1, working within Claude Science, a platform scientists can pay to use.
Okay, billions of dollars have been poured into these agentic LMs, right? Each training run to get the next increment is costing millions of dollars?
This feels like …
It critiques the hidden costs of agentic compute and the implications of companies gating scientific progress behind subscription models.
Generated by an LLM from the article and thread. Check the sources before relying on it.