Safety
154d agoNew Chain‑of‑Thought Safety Tasks Released
A new suite of nine chain‑of‑thought interpretation tasks has been released by researchers from the AI Alignment Forum. These benchmarks challenge models to predict actions, assess interventions, and analyze distributional shifts, pushing the field toward more reliable safety evaluations.
The Signal
The open‑source set invites worldwide teams to test and refine their safety methods.