“Sebastian Malaby suggests Demis Hassabis's deep background in neuroscience led him to favor reinforcement learning and underestimate the potential of scaling language models on internet data as a path to AGI.”