Demystifying Long Chain-of-Thought Reasoning in LLMs This paper investigates the emergence of long chain-of-thought (CoT) reasoning in LLMs, focusing on factors that enable structured reasoning strategies like backtracking and error correction. It analyzes the role of supervised