Training-Free Adaptation of New-Generation LLMs using Legacy Clinical Models
Sasha Ronaghi, Chloe Stanwyck, Asad Aali, Amir Ronaghi, Miguel Fuentes, Tina Hernandez-Boussard, Emily Alsentzer · 2026
Read Training-Free Adaptation of New-Generation LLMs using Legacy Clinical Models by Sasha Ronaghi, Chloe Stanwyck, Asad Aali, Amir Ronaghi, Miguel Fuentes, Tina Hernandez-Boussard, Emily Alsentzer (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/arxiv-2601-03423.txt.
Description
Sasha Ronaghi and six co-authors propose a way to adapt new language models for medical work without retraining them. Their method, called CAPT, pairs a new general model with an older clinical model that was already trained on health data. The two models can even use different vocabularies, which older pairing methods could not handle well.
CAPT uses contrastive decoding, a way of comparing model outputs, to add clinical signals while keeping the general model's fluency and reasoning. On six clinical tasks it beat both models alone and other ensemble methods. Token-level checks and doctor case studies showed more useful clinical language and fewer context errors. Hospitals with limited computing could gain from models they already own.
Catalog record
| CREATOR | Sasha Ronaghi, Chloe Stanwyck, Asad Aali, Amir Ronaghi, Miguel Fuentes, Tina Hernandez-Boussard, Emily Alsentzer |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 42 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2601-03423