As we increasingly automate AI R&D, can we also automate AI Safety research?
We built TASTE (The AI Safety Taste Evaluation), a benchmark to measure whether models can predict which research proposals experienced AI safety researchers prefer.
Show more
