Supervised Fine-tuning with Synthetic Rationale Data Hurts Real-World Disease Prediction
Why training LLMs on synthetic rationales fails to improve real-world clinical disease prediction
Supervised fine-tuning with synthetic rationale data is widely assumed to improve language model performance on clinical prediction tasks by teaching models not just what to predict but why. We test this assumption on five-year Alzheimer's disease and related dementias (ADRD) pre...
ArXiv CS.AI
Source
Generated insight by OnAirToday AI
This research reveals a critical conflict between narrative plausibility and clinical accuracy, suggesting that synthetic rationales can ironically degrade a model's predictive precision. The findings challenge the industry's reliance on chain-of-thought fine-tuning for high-stakes domains where traditional discriminative tasks are paramount.
Comments
Keep comments relevant and respectful. No links, advertising, self-promotion, or spam.