Across 504 configurations on five-year ADRD prediction, rationale-based supervised fine-tuning consistently degrades performance relative to label-only fine-tuning, despite high-quality rationales validated by experts.
and Tejada-Vera, Betzaida , title =
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.AI 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Supervised Fine-tuning with Synthetic Rationale Data Hurts Real-World Disease Prediction
Across 504 configurations on five-year ADRD prediction, rationale-based supervised fine-tuning consistently degrades performance relative to label-only fine-tuning, despite high-quality rationales validated by experts.