disease-prediction

This research paper explores the use of 24-hour wrist movement data to study and predict human health conditions and diseases, involving interdisciplinary collaborations across multiple universities.

GoodfireAI used interpretability to predict which genetic variants cause disease, validated predictions with real-world data, and released an open-source database for all variants in the NIH's ClinVar database.

This paper demonstrates that supervised fine-tuning with synthetic rationale data consistently harms prediction performance for Alzheimer's disease detection compared to label-only fine-tuning, across many configurations and model families. The degradation persists despite high-quality rationales and is attributed to a conflict between narrative plausibility and discriminative optimization.