Reproduce “Large Language Models are Few-Shot Clinical Information Extractors” 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
Monica Agrawal, Stefan Hegselmann, Hunter Lang, Yoon Kim, David Sontag
The authors test whether prompted language models can extract structured clinical information by following task instructions and a one-shot example, then using a resolver to map generated text into labels and links. On trial-arm identification, zero-shot GPT-3 reaches 0.85 accuracy across 20 abstracts, while its token-labeling score trails PubMedBERT-CRF. This small, manually assessed result does not establish broad superiority.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.