Lower Bounds for Chain-of-Thought Reasoning in Hard-Attention Transformers

Publication
arXiv preprint