Open-source "BootLoops" harness supports AI models in performing precise scientific calculations
The Decoder
BootLoops is an open-source harness built by Harvard physicist Matthew Schwartz that points language models at exact scientific calculations, filling gaps between fields instead of mimicking a human researcher. Over three months his team produced 36 manuscripts across 18 fields, and he cautions that the models often declare victory early, so humans must verify the results.
Summary written with AI by HowTechMoves, based on reporting by The Decoder.