Chain-of-thought (CoT) reasoning is commonly improved through external mechanisms such as prompting, sampling, and decoding, while the internal units shaping reasoning trajectories remain less understood. We study CoT from the perspective of reasoning trajectory structure and find that incorrect trajectories are more redundant and repetitive, whereas correct ones are more lexically diverse and information-dense. Motivated by this observation, we propose Contrastive Neuron Regulation (CNR), a sparse, backbone-frozen framework for neuron-level regulation of CoT reasoning. CNR constructs correctness-controlled expanded–concise pairs of correct trajectories, identifies MLP neurons associated with the expansion contrast, and sparsely suppresses them during generation. We consider both a fixed-coefficient inference-time intervention and an adaptive variant that learns neuron-specific suppression coefficients on a subject-specific development split while keeping the backbone frozen. Across two reasoning models and seven MATH subjects, CNR improves macro-average accuracy over the no-intervention baseline by 2.6 and 3.2 percentage points, respectively, while intervening on less than 1% of MLP neurons. The code is available at https://github.com/RUCAIBox/Contrastive-Neuron-Regulation.

Large reasoning models (LRMs) are typically trained using reinforcement learning with verifiable reward (RLVR) to enhance their reasoning abilities. In this paradigm, policies are updated using both positive and negative self-generated rollouts, which correspond to distinct sample polarities. In this paper, we provide a systematic investigation into how these sample polarities affect RLVR training dynamics and behaviors. We find that positive samples sharpen existing correct reasoning patterns, while negative samples encourage exploration of new reasoning paths. We further explore how adjusting the advantage values of positive and negative samples at both the polarity level and the token level affects RLVR training. Based on these insights, we propose an Adaptive and Asymmetric token-level Advantage shaping method for Policy Optimization, namely A3PO, that more precisely allocates advantage signals to key tokens across different polarities. Experiments across five reasoning benchmarks demonstrate the effectiveness of our approach.

Zero-shot in-context learning (ZS-ICL) aims to conduct in-context learning (ICL) without using human-annotated demonstrations.Existing ZS-ICL methods either use large language models (LLMs) to generate (input, label) pairs as pseudo-demonstrations or leverage historical pseudo-demonstrations to help solve the current problem.They assume that all problems are from the same task and traverse them in a random order.However, in real-world scenarios, problems usually come from diverse tasks, and only a few belong to the same task.The random traversing order may generate unreliable pseudo-demonstrations and lead to error accumulation.To address this problem, we reformulate ZS-ICL as a planning problem and propose a Demonstration-AWare MoNte Carlo Tree Search (MCTS) approach (DAWN-ICL), which leverages MCTS to strategically plan the problem-solving trajectories for ZS-ICL.In addition, to achieve effective and efficient Q value estimation, we propose a demonstration-aware Q-value function and use it to enhance the selection phase and accelerate the expansion and simulation phases in MCTS.Extensive experiments demonstrate the effectiveness and efficiency of DAWN-ICL on in-domain and cross-domain scenarios, and it even outperforms ICL using human-annotated demonstrations.The code is available at https://github.com/txy77/MCTS4ZSICL.

Recent advancements in long chain-of-thoughts (long CoTs) have significantly improved the reasoning capabilities of large language models (LLMs). Existing work finds that the capability of long CoT reasoning can be efficiently elicited by tuning on only a few examples and can easily transfer to other tasks. This motivates us to investigate whether long CoT reasoning is a general capability for LLMs. In this work, we conduct an empirical analysis for this question from the perspective of representation. We find that LLMs do encode long CoT reasoning as a general capability, with a clear distinction from vanilla CoTs. Furthermore, domain-specific representations are also required for the effective transfer of long CoT reasoning. Inspired by these findings, we propose GLORE, a novel representation engineering method to unleash the general long CoT reasoning capabilities of LLMs. Extensive experiments demonstrate the effectiveness and efficiency of GLORE in both in-domain and cross-domain scenarios. The code is available at https://github.com/txy77/GLoRE.