AI 速览：SCARL 通过按响应组内相对置信度与结果之间的对齐差距重塑优势值，使强化学习策略的置信度与奖励信号对齐，并缓解熵坍缩。
Reinforcement learning with verifiable rewards (RLVR) improves language-model reasoning by scoring sampled responses with a rule-based verifier and reinforcing those that pass. Where a classifier's training couples predicted probability to labels, nothing ties an RLVR policy's confidence in a response to its reward: aligned training would drive responses with the same advantage toward equal confidence, yet the advantage is blind to confidence. Under this sequence-confidence misalignment, a success the policy barely found earns the same update as one it produces with near certainty, and a confident error is corrected no harder than an exploratory one. We propose SCARL (Sequence Confidence-Aligned Reinforcement Learning), which reshapes, but never re-signs, each advantage with the alignment gap between the response's group-relative confidence and the outcome its advantage certifies, normalized into shares of each class's advantage budget. Misaligned samples absorb the force, responses with the same advantage are driven toward equal confidence, and repairing the misalignment doubles as a remedy for the entropy collapse that erodes exploration. SCARL provably reduces to its base estimator when a class's confidences agree, adds no extra passes and one hyperparameter. On Qwen3-4B-Base it lifts average Pass@32 by 2.4 points over DAPO and is best among nine compute-matched methods on both coverage and per-draw accuracy; applied unchanged across four estimators and four backbones, it improves 14 of 16 settings while preserving policy entropy and widening the confidence-correctness separation. Our code is available at https://github.com/kezhh163/scarl.
Haocheng Mei, Yuchuan Tian, Ye Yuan, Enhua Wu, Chao Xu, Allen Chen, Kai Han, Yunhe Wang
JEV decision models pair a language model backbone with a lightweight scoring head to select among natural-language options. Direct selection compares candidate answers within one request, whereas entailment reranking evaluates each candidate in a separate natural language inference (NLI) request over fixed relation options. We examine whether the readout and adaptation choices used for direct selection remain effective for cross-request reranking. We combine a mean-span readout with selective backbone adaptation. Mean-span pooling replaces the option-ending pointer with an average of the contextualized states inside each NLI option, without adding head parameters. Selective adaptation retains low-rank updates in attention and gated DeltaNet projections while removing feed-forward adapters. On MC11, which averages accuracy over eleven reranking configurations from seven sources, selective adaptation improves NeoHorse-1-4B accuracy over the all-module baseline while cutting trainable parameters by more than half, and mean-span pooling yields further consistent gains at the same parameter count.
Shaoning Sun, Mingzhu Cai, Huang He, Bingjin Chen Baidu, Siqi Bao, Yujiu Yang Tsinghua University, Hua Wu, Haifeng Wang Baidu
Language model families exhibit striking disparity in their capacity to benefit from reinforcement learning: under identical training, models like Qwen achieve substantial gains, while others like Llama yield limited improvements. Complementing data-centric approaches, we reveal that this disparity reflects a hidden structural property: distributional clarity in probability space. Through a three-stage analysis—from phenomenon to mechanism to interpretation—we uncover that RL-friendly models exhibit intra-class compactness and inter-class separation in their probability assignments to correct vs. incorrect responses. We quantify this clarity using the Silhouette Coefficient (S) and demonstrate that (1) high S correlates strongly with RL performance; (2) low S is associated with severe logic errors and reasoning instability. To confirm this property, we introduce a Silhouette-Aware Reweighting strategy that prioritizes low-S samples during training. Experiments across six mathematical benchmarks show consistent improvements across all model families, with gains up to 5.9 points on AIME24. Our work establishes distributional clarity as a fundamental, trainable property underlying RL-Friendliness.
尚无 AI 解读，展示原文摘要
cs.CL
Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表
Yilong Chen, Junyuan Shang Baidu, Yuchen Feng, Zhenyu Zhang Baidu Inc., Naibin Gu, Ziqi Wang, Tingwen Liu Institute of Information Engineering, Chinese Academy of Sciences, Shuohuan Wang Baidu 等 11 位作者
Mixture-of-Experts (MoE) is a cornerstone for scaling LLMs, yet its training dynamics remain poorly understood, often leading to sub-optimal specialization. Moving beyond static routing, we present a systematic study of the MoE lifecycle using Helmholtz Free Energyand Router Entropy. We identify a universal Three-Stage Phase Transition—Exploration, Symmetry Breaking, and Stabilization—marked by an Energy Climb and Plateau. This reflects Frustrated Exploration, caused by structural interference between specialization drives and uniformity constraints. To address this, we propose Uncertainty-Aware Routing (UAR), which aligns routing with the model’s epistemic state via: (1) Evidence-Triggered Expansion, increasing active experts for high-energy tokens, and (2) Epistemic Masking, applying load-balancing only in high-uncertainty regimes to shield mature experts. Experiments confirm UAR reduces perplexity and improves expert distinctiveness, offering a principled path toward thermodynamically aligned computation.
尚无 AI 解读，展示原文摘要
cs.CL
Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表
Shuaiyi Nie Institute of Information Engineering, Chinese Academy of Sciences, Dingsiyu Baidu, Wenyuan Zhang, Linhao Yu, Tianmeng Yang ERNIE Team, Baidu, Yao Chen, Weichong Yin, Yu Sun Baidu 等 10 位作者
Large reasoning models trained with reinforcement learning and verifiable rewards (RLVR) achieve strong performance on complex reasoning tasks, yet often overthink, generating redundant reasoning without performance gains. Existing trajectory-level length penalties often fail to effectively shorten reasoning length and degrade accuracy, as they uniformly treat all reasoning steps and lack fine-grained signals to distinguish redundancy from necessity. Meanwhile, process-supervised methods are typically resource-intensive and suffer from inaccurate credit assignment. To address these issues, we propose ATTNPO, a low-overhead process-supervised RL framework that leverages the model’s intrinsic attention signals for step-level credit assignment. We first identify a set of special attention heads that naturally focus on essential steps while suppressing redundant ones. By leveraging the attention scores of these heads, We then employ two sub-strategies to mitigate overthinking by discouraging redundant steps while preserving accuracy by reducing penalties on essential steps. Experimental results show that ATTNPO substantially reduces reasoning length while significantly improving performance across 9 benchmarks.
尚无 AI 解读，展示原文摘要
cs.CL
Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表
Yuhao Wang Nanyang Technological University, Ruiyang Ren, Yucheng Wang, Xin Zhao Renmin University of China, Jing Liu Baidu, Hua Wu, Haifeng Wang Baidu
Long-form question answering (LFQA) requires open-ended long-form responses that synthesize coherent, factually grounded content from multi-source evidence. This makes reinforcement learning (RL) reward design critical. The reward must be verifiable for faithful grounding and stable optimization. However, many standard rewards assume a unique target with an exact-match notion of correctness, which fits short-form QA and math but breaks in LFQA. As a result, current RAG systems still lack verifiable reward mechanisms, yielding unstable feedback signals and suboptimal optimization outcomes. We propose RioRAG, a framework for reinforced verifiable informativeness optimization. First, it defines informativeness as a measurable and externally verifiable objective for RL. Second, RioRAG uses nugget-centric verification with cross-source checks to enable self-evolution of smaller LLMs and to provide denser, action-discriminative rewards that mitigate reward sparsity and stabilize optimization. This formulation avoids handcrafted supervision for the policy model and strong teacher-model distillation, relying instead on externally verifiable feedback. Experiments on LongFact and RAGChecker show that RioRAG achieves higher factual recall and faithfulness, establishing verifiable reward modeling as a foundation for trustworthy long-form RAG. Our codes are available at https://github.com/RUCAIBox/RioRAG.
尚无 AI 解读，展示原文摘要
cs.CL
Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表