探索全球前沿学术脉络

01.

arXiv (CS.CL) 2026-06-15 DOI: arXiv:2606.13931

DLawBench: Evaluating LLMs Through Multi-Turn Legal Consultation

作者:

Li Zhang↗Yuzhen Shi↗Yiran Hu↗Jingwen Zhang↗Wenbo Lv↗Yubo Ma↗Wei Wang↗Rongyao Shi↗Yuanyang Qiu↗Xinran Xu↗Yuemeng Qi↗Linlin Miao↗…

Lawyer-client consultation is a critical starting point for legal services. Effective legal assistance hinges on eliciting sufficient and truthful information from clients in order to devise strategies that best protect their interests. This task requires Large Language Models (LLMs) not only to perform robust legal reasoning, but also to strategically elicit material facts through multi-turn interactions and effectively guide clients with diverse personalities. Yet existing legal benchmarks overlook this interactive capability. To fill this gap, we introduce DLawBench, a diagnostic benchmark for real-world legal consultation. Drawing on realistic client behavior, we characterize lawyer-client interactions into four types: Cooperative, Dependent, Withdrawn, and Adversarial. Using dialogues grounded in real cases, DLawBench evaluates whether LLMs can effectively conduct legal consultation under realistic conditions. DLawBench comprises 461 cases from Chinese and U.S. law, 5,532 paired fact entries, 3,411 inquiry rubrics, and 3,348 issue-resolution rubrics, and evaluates 26 representative LLMs. Systematic experiments show substantial headroom: the best-performing model, GPT-5.5, achieves only 0.562 on consultation-grounded legal reasoning. More importantly, DLawBench exposes both sycophancy in legal consultation and a paradox: models perform worse when clients need guidance most.

阅读与讨论 → 访问原文 →

02.

arXiv (CS.CL) 2026-06-16 DOI: arXiv:2606.16409

PathRouter: Aligning Rewards with Retrieval Quality in Agentic Graph Retrieval-Augmented Generation

作者:

Bo Wang↗Heyan Huang↗Yaolin Li↗Wei Tang↗Yuan Zhang↗Wenbo Li↗Mingze Gao↗Ge Shi↗Chong Feng↗

Agentic GraphRAG trains language-model agents to iteratively retrieve and reason over graph-structured evidence, enabling more accurate and context-aware decision-making by efficiently navigating complex information networks. However, outcome-only reinforcement learning suffers from answer-path reward aliasing, where correct answers may come from shortcuts rather than useful evidence paths. It also exhibits search-update ambiguity, as scalar trajectory-level feedback does not indicate which retrieval actions to adjust. To mitigate these shortcomings, we present PathRouter, a path-aware training framework for agentic GraphRAG. PathRouter jointly evaluates each trajectory along answer correctness and evidence-path overlap, yielding four trajectory categories with differentiated GRPO advantage scaling that suppresses shortcut reinforcement while preserving evidence-seeking behavior. For evidence-poor trajectories, a frozen gold-evidence teacher provides token-level KL guidance on reasoning and search-query tokens, excluding answer tokens to avoid direct response imitation. Experiments on six QA benchmarks across three model sizes show that PathRouter consistently improves answer F1 and evidence-path overlap, achieving average F1 gains of 3.1 on 3B and 4.9 on 7B models compared to a strong baseline.

阅读与讨论 → 访问原文 →

03.

arXiv (CS.AI) 2026-06-11 DOI: arXiv:2606.11265

When Poison Fails After Retrieval: Revisiting Corpus Poisoning under Chunking and Reranking Pipelines

作者:

Xi Nie↗Hongwei Li↗Shenghao Wu↗Mingxuan Li↗Jiachen Li↗Wenbo Jiang↗

arXiv:2606.11265v1 Announce Type: cross Abstract: Retrieval-Augmented Generation (RAG) systems are vulnerable to corpus poisoning attacks that manipulate downstream model outputs through malicious knowledge injection. Existing studies mainly evaluate poisoning under simplified retrieval settings, overlooking practical RAG pipelines involving document chunking, dense retrieval, reranking, and grounded generation. In this paper, we revisit corpus poisoning under realistic multi-stage retrieval pipelines and show that many existing attacks substantially degrade after reranking despite achieving high retrieval-stage relevance. We identify retrieval granularity mismatch as a key reason for this failure: document-level adversarial signals are often fragmented during chunking, while rerankers favor locally coherent and answer-bearing passages rather than globally optimized semantic similarity. Based on this observation, we propose Chunk-aware and Rerank-Consistent Poisoning (CRCP), a poisoning framework that jointly optimizes retrieval relevance, reranker consistency, and chunk-boundary robustness. CRCP explicitly models chunking transformations during optimization to generate locally self-contained adversarial passages that remain effective under varying chunking configurations. Experiments on standard RAG benchmarks with multiple retrievers and rerankers show that existing poisoning methods are highly sensitive to chunk size and reranking strategies, whereas CRCP achieves substantially higher attack success rates and stronger robustness across realistic retrieval pipelines. Our findings highlight an important realism gap in current RAG security evaluation and suggest that poisoning in modern RAG systems should be studied as a multi-stage retrieval consistency problem rather than a retrieval-only problem.

阅读与讨论 → 访问原文 →

04.

arXiv (CS.AI) 2026-06-16 DOI: arXiv:2512.22560

RollArt: Disaggregated Multi-Task Agentic RL Training at Scale

作者:

Wei Gao↗Yuheng Zhao↗Tianyuan Wu↗Shaopan Xiong↗Weixun Wang↗Dakai An↗Lunxi Cao↗Dilxat Muhtar↗Zichen Liu↗Haizhou Zhao↗Ju Huang↗Siran Yang↗…

arXiv:2512.22560v2 Announce Type: replace-cross Abstract: Agentic Reinforcement Learning (RL) trains LLMs through multi-turn interactions with environments, producing workloads that mix compute-bound prefill, bandwidth-bound decoding, CPU-heavy environment execution, and bursty reward evaluation. Existing systems either colocate all stages on a single GPU cluster or decouple them only at a coarse granularity, overlooking hardware heterogeneity and incurring substantial synchronization overhead across stages. We present ROLLART, a system for multi-task agentic RL on disaggregated infrastructure. ROLLART maps each pipeline stage to best-fit hardware, routing prefill-heavy tasks to compute-optimized GPUs, decode-heavy tasks to bandwidth-optimized GPUs, and environments to CPU clusters. It decouples rollout at the trajectory level, allowing generation, environment interaction, and reward scoring to proceed independently, so that slow or failed environments never block the others. ROLLART offloads stateless reward computation to serverless infrastructure and overlaps rollout with training via staleness-bounded asynchronous weight synchronization. Our results demonstrate that ROLLART effectively improves training throughput and achieves 1.31–2.05 \(\times\) training time reduction compared to various RL systems. We also evaluated ROLLART by training a hundreds-of-billions-parameter MoE model for Qoder product on an Alibaba cluster with above 3,000 GPUs, demonstrating its stability and scalability.

阅读与讨论 → 访问原文 →

05.

arXiv (CS.CL) 2026-06-16 DOI: arXiv:2606.15079

Ling and Ring 2.6 Technical Report: Efficient and Instant Agentic Intelligence at Trillion-Parameter Scale

作者:

Ang Li↗Ben Liu↗Bin Han↗Bin Hu↗Bin Jing↗Binbin Hu↗Bing Li↗Cai Chen↗Caizhi Tang↗Changxin Tian↗Chao Huang↗Chao Zhang↗…

Efficient and scalable agentic intelligence requires models that can deliver both low-latency responses and strong reasoning capabilities while remaining practical to train, serve, and deploy. In this report, we present Ling-2.6 and Ring-2.6, a family of models designed to address this challenge at scale. Ling-2.6 is optimized for instant response generation and high capability per output token, whereas Ring-2.6 is tailored for deeper reasoning and more advanced agentic workflows. Instead of training from scratch, we upgrade the Ling-2.0 base model through architectural migration pre-training and large-scale post-training. This upgrade is guided by a unified co-design of model architecture, optimization objectives, serving systems, and agent training environments, enabling improvements in both model capability and deployment efficiency. At the architectural level, we introduce a hybrid linear attention design that integrates Lightning Attention with MLA, improving the efficiency of long-context training and decoding. To further enhance token efficiency, we optimize capability per output token through Evolutionary Chain-of-Thought, Linguistic Unit Policy Optimization, bidirectional preference alignment, and shortest-correct-response distillation. For agentic capabilities, we propose KPop, a reinforcement learning framework designed to support stable training of Ring-2.6-1T on large-scale environment-grounded data. KPop improves training efficiency through asynchronous scheduling across coding, search, tool use, and workflow execution, enabling scalable learning from complex agent-environment interactions. Together, Ling-2.6 and Ring-2.6 provide a practical pathway toward efficient, scalable, and open agentic systems. We open-source all checkpoints in the 2.6 family to support further research and development in practical agentic intelligence.

阅读与讨论 → 访问原文 →

06.

arXiv (CS.AI) 2026-06-16 DOI: arXiv:2602.12670

SkillsBench: Benchmarking How Well Agent Skills Work Across Diverse Tasks

作者:

Xiangyi Li↗Yimin Liu↗Wenbo Chen↗Bingran You↗Zonglin Di↗Yifeng He↗Shenghan Zheng↗Kyoung Whan Choe↗Jiankai Sun↗Shuyi Wang↗Chujun Tao↗Binxu Li↗…

arXiv:2602.12670v4 Announce Type: replace Abstract: Agent Skills are structured packages of procedural knowledge that augment large language model (LLM) agents at inference time. Despite rapid adoption, there is no standard way to measure whether they actually help. We present SkillsBench, a benchmark whose current inventory contains 87 tasks across 8 domains paired with curated Skills and deterministic verifiers. Our latest aggregate evaluation runs the 87-task benchmark under matched no-Skills and curated-Skills conditions for 18 model-harness configurations. Curated Skills raise the average pass rate from 33.9% to 50.5% (+16.6 percentage points; 25.5% normalized gain), with configuration-level gains ranging from +4.1 to +25.7 pp. Focused Skills with at most three modules outperform larger or exhaustive bundles, and smaller models with Skills can match larger models without them. SkillsBench establishes paired evaluation as the foundation for rigorous measurement of Skill efficacy on agentic, expertise-heavy work.

阅读与讨论 → 访问原文 →

07.

arXiv (CS.AI) 2026-06-16 DOI: arXiv:2505.20030

Multiple Descents in Deep Learning as a Sequence of Order-Chaos Transitions in LSTM Networks

作者:

Wenbo Wei↗Fan Xu↗Nicholas Chong Jia Le↗Choy Heng Lai↗Ling Feng↗

arXiv:2505.20030v2 Announce Type: replace-cross Abstract: We observe a novel `multiple-descent' phenomenon during the learning process of a recurrent neural network called long-short-term memory (LSTM) networks during its training on real-world task, in which the performance goes through long cycles of up and down trends multiple times after the model is overtrained. By carrying out asymptotic stability analysis of the models, we found that the cycles in performance – indicated by loss function in test data – are closely associated with the phase transition process between order and chaos of the model, and the local optimal training step are consistently at the critical transition point between the two phases. More importantly, the most optimal point of the model usually occurs at the first transition from order to chaos, where the `width' of the `edge of chaos' is often the widest, allowing the best exploration of weight configurations for learning.

阅读与讨论 → 访问原文 →

08.

arXiv (CS.AI) 2026-06-15 DOI: arXiv:2606.14078

Rethinking Backdoor Adversarial Unlearning through the Lens of Catastrophic Forgetting in Continual Learning

作者:

Zhenqian Zhu↗Yamin Hu↗Yujiang Liu↗Luping Wei↗Wenbo Hou↗Bin Li↗Haodong Li↗Wenjian Luo↗

arXiv:2606.14078v1 Announce Type: cross Abstract: Existing studies reveal that current backdoor defenses exhibit limited robustness and often fail against specific types of attacks. More concerningly, prevailing safety tuning strategies tend to provide only superficial safety protection, as they fall short of completely eliminating the backdoor effects. In this work, we present a novel formulation of backdoor learning and unlearning as a sequential, three-stage process from a continual learning perspective. Within this framework, we formally define complete backdoor unlearning and further derive the necessary conditions for achieving it based on the mechanism of catastrophic forgetting. Guided by these insights, we propose Blind Inversion-Backdoor Adversarial Unlearning (BI-BAU), which formulates the generation of adversarial examples satisfying the unlearning conditions as a blind inversion problem. We solve this by integrating the bi-level optimization process of adversarial training into an Expectation-Maximization (EM) algorithm framework to optimize the maximum a posteriori (MAP) objective. Furthermore, BI-BAU is extended to untargeted adversarial scenarios with unknown target classes, as well as to multi-modal contrastive learning tasks, enhancing its applicability to real-world deployment scenarios where pre-trained models may be compromised. Extensive experiments demonstrate that our method exhibits general applicability across a wide spectrum of backdoor attacks and can effectively and thoroughly eliminate the backdoor effects from a backdoor model.

阅读与讨论 → 访问原文 →

09.

arXiv (CS.AI) 2026-06-12 DOI: arXiv:2606.12935

MARS: Margin-Adversarial Risk-controlled Stopping for Parallel LLM Test-time Scaling

作者:

Wenbo Chen↗Puheng Li↗Mengyang Liu↗Weijie Su↗Tianpei Xie↗

arXiv:2606.12935v1 Announce Type: new Abstract: Parallel test-time scaling samples many reasoning traces and majority-votes their answers, improving LLM accuracy but requiring traces to run to completion, incurring substantial computational overhead. We observe that probing partial traces at intermediate checkpoints can extract current answers without disrupting generation, revealing an evolving aggregate vote. Based on this observation, we introduce MARS, a margin-adversarial stopping rule that estimates which active traces are likely to change their answers and stops once the leader remains safe under a conservative bound on future vote movement. The rule separates two sources of uncertainty. It learns the trace-level switch probabilities that determine how much of the current margin is likely to be retained, while handling the harder question of where switching traces land through an adversarial bound calibrated from warmup traces. With true switch probabilities, MARS guarantees with high probability that the early-stopped answer matches the full-budget vote. In practice, a five-feature logistic model closely matches oracle switching behavior. Across three reasoning models and three competition-math benchmarks, MARS saves 25-47% of self-consistency tokens and 14-29% on top of DeepConf Online, a strong confidence-weighted baseline that already filters and truncates weak traces, while matching the accuracy of the corresponding full-budget baselines.

阅读与讨论 → 访问原文 →

10.

arXiv (CS.CL) 2026-06-12 DOI: arXiv:2606.05405

Agents' Last Exam

作者:

Yiyou Sun↗Xinyang Han↗Weichen Zhang↗Yuanbo Pang↗Tianyu Wang↗Yuhan Cao↗Yixiao Huang↗Chris Duroiu↗Haoyun Zhang↗Jeffrey Lin↗Weishu Zhang↗Tyler Zeng↗…

Recent AI systems have achieved strong results on a wide range of benchmarks, yet these gains have not translated into economically meaningful deployment across many professional domains. We argue that this gap is largely an evaluation problem: widely used benchmarks lack sustained performance measurement on real and economically valuable workflows. This paper introduces Agents' Last Exam (ALE), a benchmark designed to evaluate AI agents on long horizon, economically valuable, real world tasks with verifiable outcomes. Developed in collaboration with 250+ industry experts, ALE covers non-physical industries defined with reference to O*NET / SOC 2018 (the U.S. federal occupational taxonomy). It is organized around a task taxonomy with 55 sub fields grouped into 13 industry clusters covering 1K+ tasks. Current results show that the hardest tier remains far from saturated: across mainstream harness and backbone configurations, the average full pass rate is below 1%. ALE is designed as a living benchmark: its task pool grows continuously as new workflows and industries are onboarded. More broadly, ALE is intended not merely as another leaderboard, but as an instrument for closing the gap between benchmark success and GDP relevant impact.

阅读与讨论 → 访问原文 →

11.

arXiv (CS.CL) 2026-06-16 DOI: arXiv:2606.15396

CHILLGuard: Towards Fine-Grained Chinese LLM Safety Guardrail with Scalable Data Construction and Model-aware Preference Alignment

作者:

Wenbo Yu↗Bohua Wang↗Hao Fang↗Kuofeng Gao↗Jingru Zeng↗Xiaochen Yang↗Tianyi Zhang↗Xiaoxiao Ma↗Jiawei Kong↗Hao Wu↗Bin Chen↗Shu-Tao Xia↗…

Malicious content generated from large language models (LLMs) could pose severe safety risks and ethical concerns. While existing LLM safety guardrails excel in English or multilingual settings, they lack adaptation to Chinese-specific regulatory policies, cultural context and linguistic nuances, failing to support fine-grained risk classification for diverse deployment needs. In this paper, we introduce a 5-macro, 31-micro category fine-grained risk taxonomy for Chinese scenarios, and build CHILLGuard: a dedicated Chinese LLM content safety guardrail. To address the critical scarcity of high-quality annotated Chinese safety data, we propose a scalable multi-stage data construction pipeline: we expand multi-source corpus via retrieval-augmented generation, generate implicit harmful samples through prompt engineering rewriting, and refine high-quality data via multi-model voting-based label calibration. Based on this, we build CHILLGuardTrain, a large-scale training set with 405,007 samples, and CHILLGuardTest, a rigorously curated annotated test set with 51,745 samples. We then train CHILLGuard on CHILLGuardTrain under a generator-classifier collaborative framework via Model-aware Direct Preference Optimization. Extensive experiments under multiple settings demonstrate the state-of-the-art performance of CHILLGuard, e.g., a 15.92% improvement of F1 score over Qwen3Guard-8B-Strict on our benchmark. We will release our resources at https://github.com/cswbyu/CHILLGuard.

阅读与讨论 → 访问原文 →

12.

arXiv (CS.CL) 2026-06-16 DOI: arXiv:2606.15893

BALTO: Balanced Token-Level Policy Optimization for Hallucination Mitigation

作者:

Ning Li↗Zixuan Guo↗Yan Xu↗Wenbo Fei↗Yifan Niu↗Chang Luo↗Yasheng Wang↗Weiwen Liu↗Yong Yu↗Weinan Zhang↗

Hallucinations remain a major obstacle to deploying large language models (LLMs) in knowledge-intensive settings, where generated responses must be faithfully grounded in provided evidence. Reinforcement learning (RL) is a promising direction for hallucination mitigation, but response-level faithfulness rewards suffer from a granularity mismatch: localized hallucinations can cause supported content to receive spurious penalties. Although recent work introduces fine-grained feedback such as claim-level verification and token-level rewards, unbalanced credit assignment can still induce length, verbosity, or optimization-noise biases. We propose BALTO, a Balanced Token-level Policy Optimization framework for hallucination mitigation. BALTO extracts checkable factual claims, verifies them against the reference context, and projects claim-level judgments to token-level labels. A balanced token-level credit assignment mechanism is introduced into the framework. This design redistributes probability mass from unsupported content toward faithful content, rather than suppressing the entire response. We systematically analyze the limitations of response-level rewards from a theoretical standpoint, and prove BALTO's advantages in training stability and optimization efficiency for hallucination mitigation. Experiments on ConFiQA, RAGTruth, and FinLLM-Eval show that BALTO achieves the highest faithfulness across all six model–benchmark settings and consistently outperforms existing post-training baselines in Q-Score, demonstrating a stronger faithfulness–informativeness trade-off.

阅读与讨论 → 访问原文 →