Machine Learning and Inner Alignment

Curet: I'd like to change topics again a little and talk about inner alignment. So in the paper, you describe inner alignment and is base leve the task of making sure that the base objective is the same as the masa objective. And you give a taxonomy of ways that a inner alignment can fail. There's proxy alignment, where the thing that am yeur er er er specifically, that inner alinement can fail that, but like, you don't know that it's failed. But if inner linement fails and like, the system just does something totally different, then normal machine learning willhopfully take care of that. Um. Thirdly, sub optimality