Launch a whole new initiative to produce guidance and benchmarks for assessing AI abilities, by using a focus on abilities that might bring about harm. This setup may be more Value-efficient, since it reduces the https://aistudio.ml