Claude-Shaped Science
In this guest post, Prof. Matthew Schwartz returns after “Vibe Physics,” where he discussed similarities in capability between Claude and a physics graduate student, to describe a new approach to AI-accelerated science. Here he explains what happened when he stopped fighting Claude and allowed it to find “Claude-shaped” problems—ones best suited to the capabilities of current LLM tools. This led him to build BootLoops, a toolkit for exact calculations in quantitative science.
Schwartz notes that agentic AI is improving rapidly and models seem smarter: they know more, make fewer mistakes, and have better ideas. Trend lines make it easy to speak confidently about AI revolutionizing science, but academic scientists often feel a disconnect—models are solving challenging, longstanding problems, yet mostly in well-scoped applications of existing techniques. Many headlines are in mathematics, where a problem can be stated completely and an answer checked absolutely, but most of science is not like that, leaving researchers and students frustrated and anxious.
The core conflict, as he sees it, is that although these models are brilliant, working like a human scientist is not what current LLMs do best. Claude and GPT are good at science but are not scientists: they are smart, but it can take a lot of hand-holding to get them to produce anything of scientific value. Physicists would call this an “impedance mismatch”—two systems that each work fine but are poorly matched, so most of what one puts in never gets through to the other—between what scientists want and what AI does well.
He started looking for examples where the impedance mismatch is less acute. He began by having Claude build an accessible suite of tools for mathematical physics, and before long the tools found uses for other problems. This iterative process generated a set of software and scientific protocols he calls BootLoops, which functions as a harness for the LLM, much like Claude Code or Claude Science is a harness for Claude, or Codex is a harness for GPT. He finds BootLoops especially well-suited for a class of quantitative problems in science.
Because similar calculations often turn up across very disparate areas of science, Claude found connections to ecology, population genetics, and a dozen other fields. These connections were often technically correct but scientifically unremarkable at first, so Schwartz worked with domain experts to steer BootLoops toward questions those fields care about. BootLoops is also open-source, so it can be used with whatever model you like.