AnchorSIPS: A Synthetic Dataset and Evaluation Resource for Evidence-Supported Psychosis-Risk Symptom Measurement
The episode discusses AnchorSIPS, a synthetic dataset of ten thousand psychosis-risk interviews designed to test AI systems. Hosts analyze findings showing that while current large language models can predict diagnoses, they fail at citing specific evidence from the transcript to support their claims. The resource aims to measure true evidence-supported reasoning.