Claude-Shaped Science

Prof. Matthew Schwartz describes a new approach to AI-accelerated science called BootLoops, a toolkit designed around "Claude-shaped" problems—quantitative calculations that leverage Claude's strengths rather than forcing it to work like a traditional scientist. By accepting what Claude does best rather than fighting its limitations, Schwartz built an open-source harness that has uncovered connections across disparate fields like ecology and population genetics, demonstrating how tailoring AI applications to match LLM capabilities can produce scientifically valuable results.