The unexpected capacities of Large Language Models to understand concepts and engage in convincing dialogue, including the use of theory of mind, is remarkable, given that these models are at bottom nothing but sequence learners. In this talk, I’ll relate recent developments in large unsupervised foundation models with both “society of mind” models of cognition, and with theories of intelligence that place our evolutionary development and that of other large-brained animals in a social framework. Some implications of this emerging multiscale social picture of cognition will be sketched.