Patient message response drafting on IPPM
15Content-Level PrecisionSFT
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| SFTadaptation=Supervised Fine-Tuning, aggregation=Average across 3 LLMs2026.01 | 15 | 16 | 14 | 64 | 57 | 60 | |
| TADPOLEadaptation=TADPOLE, aggregation=Average across 3 LLMs2026.01 | 13 | 18 | 14 | 54 | 65 | 59 | |
| RAGadaptation=Retrieval-Augmented Generation, aggregation=Average across 3 LLMs2026.01 | 11 | 30 | 13 | 48 | 66 | 56 | |
| 0-Shotadaptation=None, aggregation=Average across 3 LLMs2026.01 | 7 | 26 | 10 | 49 | 74 | 58 | |
| Themeadaptation=Thematic Prompting, aggregation=Average across 3 LLMs2026.01 | 6 | 30 | 9 | 47 | 80 | 58 |