Harvard physicist Matthew Schwartz has developed "BootLoops," an open-source harness that utilizes language models for precise scientific calculations and bridging knowledge gaps. Built during a stint as a visiting researcher at Anthropic, the harness leverages models like Claude to find connections and solve complex problems across diverse disciplines.
Using BootLoops, Schwartz and 19 co-authors produced 36 manuscripts across 18 fields in just three months, tackling everything from particle physics and ecology to population genetics and economics. Notable achievements include computing fifteen previously unsolved integrals in physics and solving a 20-year-old equation in neutral biodiversity theory. Despite these rapid advancements, Schwartz warns that models frequently declare victory prematurely or draw incorrect conclusions from accurate math, meaning human oversight and domain expertise remain essential.
This open-source tool highlights how AI is rapidly transforming academic research and automating technical tasks that once took years. Why it matters: BootLoops demonstrates that AI can accelerate multi-disciplinary scientific discovery, though human verification is still required to ensure accuracy.