Explore the computational power and learning dynamics of iterated sequence-to-sequence models in this 54-minute lecture by Nati Srerbo from Toyota Technological Institute at Chicago. Delve into the concept of transformers as a computational model, focusing on systems that apply the same function at each step to generate the next token. Examine the expressive capabilities of these models, even with simple base classes, and investigate the sample and computational complexity of learning processes. Gain insights into both end-to-end learning and approaches that leverage the entire "chain of thought" in model training and inference.