BootLoops is for scientific calculations

The Decoder reports that Harvard physicist Matthew Schwartz built BootLoops, an open-source harness that uses language models for exact scientific calculations and is meant to bridge gaps between fields. He is also a visiting researcher at Anthropic. The source line names Anthropic. This piece is The Decoder’s account. Schwartz says he stopped trying to use Claude like a human researcher and started looking for Claude-shaped problems: work scientists want to do that current models can actually do.

Over three months, the team and 19 co-authors produced 36 manuscripts across 18 fields, from particle-physics calculations to ecological modeling and genome analysis. In particle physics, Claude computed 30 integrals with BootLoops, fifteen reproducing known results and fifteen computed for the first time. In ecology, it solved a 20-year-old equation from neutral biodiversity theory that had been impossible to compute at scale. Applied to data, tree species on Barro Colorado Island are changing 4.5 times faster than the theory allows. Ecologist James O'Dwyer then helped turn that finding into a better predictive model.

Schwartz warns that the models often declare victory too early, or draw wrong conclusions from correct calculations, so human oversight and verification remain necessary. He says automated checks are not reliable. The model also gravitates toward old, heavily cited debates rather than genuinely new questions. The projects were heavy on compute and tokens. He now calls a Python-for-engineers course unnecessary, where two years ago he would have called it essential, because Claude can do those assignments. That is his judgment, not a notice that a course has been cancelled.