Reproduce “EXCISE: Locating and Excising Distracting Sentences in Medical Question Answering” 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
Yerim Hong, Sohyun Son, Heeseung Chung, Mujeen Sung
EXCISE tests which sentence shifts a medical multiple-choice model’s answer, then blocks it only when its representation also appears out of place. On MEDQA-TD, it raises Llama-3.1-8B-Instruct accuracy from 40.7% to 57.4%, while clean MedQA accuracy falls from 61.7% to 59.6%. These results concern controlled, synthetic distractors, not clinical use.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.