Matthew Schwartz releases BootLoops for precise scientific calculations with language models
Physicist Matthew Schwartz released the source code for BootLoops, a tool for scientific calculations with language models. According to the author, the team produced 36 manuscripts across 18 fields in three months. However, the results required review and guidance from domain experts.
Physicist Matthew Schwartz from Harvard University created and released BootLoops, a tool for precise scientific calculations with language models. The source code is available on GitHub. The tool is also intended to help find connections between different scientific fields.
According to the author, the team with 19 co-authors produced 36 manuscripts across 18 fields in three months. With the help of BootLoops, Claude models calculated 30 integrals over several weeks: 15 reproduced known results, and another 15 were calculated for the first time, according to the author. The scientific significance of the results often emerged only after domain experts became involved and determined the next direction of the work.
Schwartz warns that Claude models prematurely declare tasks complete and can draw incorrect conclusions even from correct calculations. According to him, automated checks are not reliable either. He also describes the projects as demanding in terms of computing resources and token consumption.
Why it matters
The released tool gives researchers an opportunity to try language models for specific scientific calculations and for finding connections between fields. However, the results described show that a correct calculation alone does not guarantee a correct interpretation. Practical use therefore requires expert review as well as resources to run the computations.
Two audiences, two different impacts
What this means
For individuals
A researcher can try BootLoops for individual calculations, but must independently verify task completion and the interpretation of the result; according to the author, models can make mistakes in both.
For a business
For organizations with research teams, using BootLoops entails demands on computing resources, token consumption and the involvement of domain experts in reviewing results.
ProcessesCheck the original
Event sources
only one source so far · 1 publisher, 1 independent. We count feeds from the same owner only once.