Researchers from Tsinghua University and Shanghai AI Lab have introduced a new AI framework called Diagram of Thought (DoT), which models iterative reasoning in large language models (LLMs) as the construction of a directed acyclic graph (DAG) within a single model. This development aims to enhance the reasoning capabilities of LLMs through mathematical rigor. Additionally, Google DeepMind has developed a multi-turn chain-of-thought online reinforcement learning approach called SCoRe to improve self-correction in LLMs using self-generated data. SCoRe has achieved state-of-the-art performance, improving metrics such as MATH by 15.6% and HumanEval by 9.1%. Another novel approach, Iteration of Thought (IoT), leverages inner dialogue for autonomous reasoning, while the Hidden Chain-of-Thought (HCoT) framework speeds up LLM inference while preserving multi-step reasoning capabilities. These advancements highlight the growing focus on improving reasoning and self-correction capabilities in LLMs.