Fine-tuning and structured prompting strategies for question answering over full-text biomedical research articles
Kaiming Tao, Rohit Satija, Jinru Zhou, Zachary A Osman, Vineet Ahluwalia, Chiara Sabatti, Robert W Shafer · 2026
Read Fine-tuning and structured prompting strategies for question answering over full-text biomedical research articles by Kaiming Tao, Rohit Satija, Jinru Zhou, Zachary A Osman, Vineet Ahluwalia, Chiara Sabatti, Robert W Shafer (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/plos-pone-0351631.txt.
Description
Kaiming Tao and colleagues tested whether coaching large language models makes them better at answering set questions from full research papers. Their test bed was HIV drug resistance studies, with sixteen fixed questions per paper about sequences, patients, and treatments. They compared three well-known models, then tried extra training on worked examples and a prompting style built from question-specific rules.
Extra training sharpened precision, meaning fewer wrong facts offered, and helped the models in different degrees. The rule-based prompts mainly lifted recall, meaning fewer facts missed, though gains were uneven. Stacking prompts on top of training added nothing further. For teams hoping to mine papers automatically, the study maps which coaching buys which kind of accuracy, in a field where a missed treatment detail can mislead real research.
Catalog record
| CREATOR | Kaiming Tao, Rohit Satija, Jinru Zhou, Zachary A Osman, Vineet Ahluwalia, Chiara Sabatti, Robert W Shafer |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 51094 |
| LICENSE | CC_BY |
| RIGHTS | CC BY |
| SOURCE | PLOS |