Archive

We ran HealthBench on our health AI's safety layer. It scored lower than the bare model.

John Alexander Kamau