Arthur Conmy announced that he is joining Anthropic to work on aligning upcoming models as they are trained. Conmy is known for mechanistic-interpretability work including automated circuit discovery, and had previously been a senior research engineer at Google DeepMind.