The Research That Can Answer Questions About AI Evaluation Composability
Before anyone can say whether an evaluation result survives fine-tuning, we need to know how much of that result is noise. Two research projects that would move the composability question forward.