Reproduce “Weight-space programming for auditing clinical decision networks” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Humza Nusrat, Omar Nusrat
The authors test whether a language model can inspect and edit weights in small synthetic head-and-neck radiotherapy planning networks to find shortcuts under a four-rule protocol. Claude Opus 5.5 detected 16 of 18 planted flaws and repaired all 16 detected cases, changing a median of three parameters. It cleared only two of five healthy networks; protocol testing found every flaw faster.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.