Open-source "BootLoops" harness supports AI models in performing precise scientific calculations
Harvard physicist Matthew Schwartz has built "BootLoops," an open-source harness that uses language models for exact scientific calculations, with source code available on GitHub. Schwartz, currently a visiting researcher at Anthropic, described the work in an Anthropic guest post.
Over three months, Schwartz and 19 co-authors produced 36 manuscripts across 18 fields, spanning particle physics calculations, ecological modeling, population genetics, economics, and linguistics. In particle physics, he said Claude computed 30 integrals using BootLoops, fifteen reproducing known results and fifteen computed for the first time. In ecology, Claude solved a 20-year-old equation from neutral biodiversity theory; applied to data, the results indicated that tree species composition on Barro Colorado Island in the Panama Canal is changing 4.5 times faster than the theory allows, and ecologist James O'Dwyer helped turn the finding into a predictive model. In population genetics, the team analyzed 5.7 billion mutation pairs from the 1000 Genomes Project and found evidence for gene conversion. Other projects included an AI data editor for economics journals that checked 4,452 replication packages, published as an NBER Working Paper, and a word stress database covering 6,072 languages built with three linguists.
Schwartz warned that the models often declare victory too early, misjudge task duration, brute-force calculations, and can reach wrong conclusions even when calculations are correct. Automated checks aren't reliable, he said, and he advised human experts to "look at everything yourself." He noted the projects were compute- and token-intensive and that models gravitate toward old, heavily cited debates. He also said AI is changing science fast enough that planning ahead is becoming nearly impossible, questioning three-year grants and PhD training, and calling a "Python for Engineers" course unnecessary today. Human guidance and taste remain indispensable, he said.
Based on reporting from the original publisher. Visit the source for full context and later updates.
Publisher excerpt
Harvard physicist Matthew Schwartz used the open-source tool BootLoops and Claude to produce 36 manuscripts across 18 fields in three months, from particle physics to linguistics. But the results often only became scientifically valuable once human experts stepped in. "Look at everything yourself," Schwartz says. The article Open-source "BootLoops" harness supports AI models in performing precise scientific calculations appeared first on The Decoder .