Open-source "BootLoops" harness supports AI models in performing precise scientific calculations
decoder ·
Harvard physicist Matthew Schwartz has developed BootLoops, an open-source harness that leverages large language models to perform precise scientific calculations and bridge disciplinary gaps. In just three months, Schwartz and 19 co-authors produced 36 manuscripts across 18 diverse fields, ranging from particle physics to ecology and linguistics. The system successfully computed previously intractable integrals in physics and solved a longstanding equation in neutral biodiversity theory, revealing faster-than-expected changes in tree species composition. It also analyzed billions of mutation pairs to find evidence for gene conversion and automated the verification of replication packages for economics journals. Schwartz argues that AI should address "Claude-shaped problems" where it can connect fragmented areas of human knowledge rather than replacing traditional research methods entirely. However, he warns that models often declare victory prematurely or draw incorrect conclusions despite accurate calculations, necessitating strict human oversight. Consequently, while AI accelerates data processing and pattern recognition, expert guidance remains indispensable for validating results and setting scientific direction.
decoder