Reproduce “Shaping & Monitoring Models through their Internal Representations” 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
Thomas Fel, Matthew Kowal, Matt Feiszli, Satchel Grant, Andrew Lee, Chris Earls, Fatih Dinc, Vasudev Shyam and 18 more
The authors introduce Mechanistic Shaping, which uses internal representations to penalize a model’s reliance on specified concepts or their expression. In a constructed, perfectly drain-confounded pneumothorax benchmark, reliance shaping raises worst-group accuracy from 33.3% with ordinary training to 61.3%, above the 56.0% group-balanced reference. This result concerns a 100-patient test, not clinical deployment; concepts may remain encoded.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.