Prof. Matthew Schwartz, a guest author for Anthropic Research, describes a new approach to AI‑accelerated science that he has called BootLoops. BootLoops is an open‑source toolkit that harnesses the capabilities of Claude, Anthropic’s large language model, to solve quantitative problems in science. The idea behind BootLoops is to identify problems that are best suited to the strengths of current LLMs, which Schwartz calls "Claude‑shaped" problems. These are problems that can be solved by a technique from mathematics, physics, or computer science, but that are not well‑suited to the way scientists typically work. By focusing on these problems, Schwartz was able to build a suite of tools that can be used across a wide range of scientific fields, including physics, ecology, population genetics, and more. The toolkit is designed to be a harness for the LLM, similar to how Claude Code or Claude Science are harnesses for Claude, or Codex is a harness for GPT. BootLoops is also open‑source, so it can be used with whatever model you like. Schwartz’s work with BootLoops has led to substantive advances in many research areas, and he has shared his thinking on how to resolve the impedance mismatch between human and AI scientists today.
Claude-shaped science
Prof. Matthew Schwartz explains how he leveraged Anthropic’s Claude to identify and solve problems best suited to current large language models, creating the open‑source BootLoops toolkit. By focusing on "Claude‑shaped" problems—tasks where Claude’s breadth, coding, and mathematical knowledge excel—he has applied BootLoops across physics, ecology, genetics, and more, collaborating with domain experts to ensure relevance.
Why it matters
Demonstrates a practical framework for aligning AI capabilities with scientific needs, potentially accelerating research across disciplines.
Topic dossiers Anthropic & Claude