Read our recent paper — a collaboration between scientists from OpenEvidence (Xyla Research Group), MIT, Harvard Medical School, The Hospital for Sick Children, and Brigham and Women’s Hospital.
In an era of extremely capable general purpose large language models (LLMs), we investigate the need for specialized clinical models, LLMs that are customized for clinical text such as patient notes and discharge summaries. In an extensive empirical analysis, we show that relatively small specialized clinical models can substantially outperform larger language models trained on general text.