Reinforcement learning (RL) is a natural fit for agentic knowledge base question answering (KBQA), where a model must issue executable actions, observe knowledge-base feedback, and eventually return an answer. However, current RL-based KBQA systems mainly optimize sparse rewards from the final answer, leaving intermediate action errors weakly supervised. This is especially limiting for logical-form annotated KBQA benchmarks: gold logical forms can be converted into executable action sequences, but existing pipelines use them mainly for warm-start data construction rather than for on-policy RL updates. We propose GAPD, a training-time Gold-Action Policy Distillation framework that adds dense token-level guidance to outcome-based RL. To align gold actions with on-policy student rollouts, GAPD uses MID-ANCHOR MATCHING: it treats the intermediate entities reached during student exploration and gold execution as state anchors, and matches student states to gold states through these explored entity sets. The current policy conditioned on this aligned gold action serves as a stop-gradient teacher, whose token distribution is distilled back to the ordinary student policy over generated action-token spans. GAPD consistently surpasses the current state of the art on WebQSP, GrailQA, and GraphQ.

LLM-based annotation has advanced rapidly, yet existing benchmarks remain misaligned with real-world practice. Actual annotation often involves lengthy, task-specific guidelines, and different annotators may produce divergent but reasonable judgments, especially on boundary cases, where single-labeled dataset fail to reflect this disagreement. This gap also limits existing single-agent and workflow-based annotation systems, which often lack adaptability to complex, evolving guidelines. To bridge this gap, we introduce ANT-A Benchmark, a real-world multi-annotator benchmark spanning 7 datasets across domains and modalities, with task-specific guidelines and 5 independent human labels per instance. ANT-A enables a distribution-aware evaluation, using Fleiss’ 𝜅 to assess how closely model annotations align with human labeling distributions. We further propose ANT-OR, a multi-agent annotation framework that refines prompts by distilling contrastive decision rules from positive and negative feedback. Experiments show that ANT-OR improves agreement@≥2 accuracy by 4.5 points over the strongest baseline and achieves the highest alignment with human annotator distributions as measured by Fleiss’ 𝜅 within 2 points. Code and benchmark are available at https://github.com/Krueske/ANT-OR.

Patient-facing medical LLMs are typically evaluated for clinical knowledge and safety, but clinically correct answers can still fail patients by missing fear, mishandling uncertainty, or offering emotionally unsafe reassurance. We introduce MedEQ, a benchmark for clinical emotional intelligence in medical communication. MedEQ includes 413 clinically grounded consultation scenarios across 51 specialties and 7,336 expert-refined rubric items over 21 core dimensions. These expert-validated rubrics operationalize emotional cue recognition, empathic adequacy, ethical emotional regulation, and affective safety, enabling diagnostic evaluation rather than a single holistic empathy score. Using an LLM-as-judge rubric pipeline, we find that strong contemporary models often handle surface compassion, safety, and response structure, yet remain weak on deeper emotional inference, personalization, mixed emotions, cultural context, and real-time responses to illness-related distress. We further conduct a proof-of-concept rubric-guided reinforcement learning study, showing that MedEQ-derived rubric signals can transfer to an independent emotional intelligence benchmark: rubric-guided training of Qwen3-32B improves EQ-Bench performance from 42.42 to 81.49 while preserving or slightly improving standard medical benchmark scores. MedEQ therefore makes clinical emotional intelligence measurable and decomposable, and provides evidence that expert-refined rubrics can support the evaluation and improvement of patient-facing medical AI.

Enhancing large language models (LLMs) with external tools has become a promising approach for solving complex tasks. As the number of available tools grows, context-based prompting methods increasingly rely on retrieval mechanisms. A common solution is to represent each tool with a unique token and train LLMs to generate the corresponding token during inference. However, this approach suffers from linear growth in representation space, leading to scalability challenges. It also limits generalization to novel or rare tools and underutilizes collaborative signals among tools in downstream tasks. In this paper, we propose SGTC, a generative tool invocation framework that introduces structure-aware semantic tokenization to encode tools as discrete code sequences. This method ensures similar tools share subtokens, enabling compression of the representation space and facilitating token sharing for new tools. We further introduce a post-guided, multistage iterative training strategy on a shared backbone model, where collaborative signals from downstream tasks guide the dynamic refinement of tool representations. Extensive experiments on the ToolBench dataset, which includes over 47,000 APIs, demonstrate the effectiveness of SGTC across various tasks, showcasing its potential as a scalable and generalizable generative tool-using paradigm in large-scale tool usage scenarios. The code is available at https://github.com/OPilgrim/Toolscaler.

Hallucination has emerged as a significant barrier to the effective application of Large Language Models (LLMs). In this work, we introduce a novel Attention-Guided SElf-Reflection (AGSER) approach for zero-shot hallucination detection in LLMs. The AGSER method utilizes attention contributions to categorize the input query into attentive and non-attentive queries. Each query is then processed separately through the LLMs, allowing us to compute consistency scores between the generated responses and the original answer. The difference between the two consistency scores serves as a hallucination estimator. In addition to its efficacy in detecting hallucinations, AGSER notably reduces computational complexity, requiring only three passes through the LLM and utilizing two sets of tokens. We have conducted extensive experiments with four widely-used LLMs across three different hallucination benchmarks, demonstrating that our approach significantly outperforms existing methods in zero-shot hallucination detection.

Large language models (LLMs) augmented with retrieval systems have significantly advanced natural language processing tasks by integrating external knowledge sources, enabling more accurate and contextually rich responses. To improve the robustness of such systems against noisy retrievals, Retrieval-Augmented Fine-Tuning (RAFT) has emerged as a widely adopted method. However, RAFT conditions models to generate answers even in the absence of reliable knowledge. This behavior undermines their reliability in high-stakes domains, where acknowledging uncertainty is critical. To address this issue, we propose Divide-Then-Align (DTA), a post-training approach designed to endow RAG systems with the ability to respond with “I don’t know” when the query is out of the knowledge boundary of both the retrieved passages and the model’s internal knowledge. DTA divides data samples into four knowledge quadrants and constructs tailored preference data for each quadrant, resulting in a curated dataset for Direct Preference Optimization (DPO). Experimental results on three benchmark datasets demonstrate that effectively balances accuracy with appropriate abstention, enhancing the reliability and trustworthiness of retrieval-augmented systems.

Knowledge graph embedding (KGE) models provide a low-dimensional representation of knowledge graphs in continuous vector spaces. This representation learning enables different downstream AI tasks such as link prediction for graph completion. However, most embedding models are only designed considering the algebra and geometry of the entity embedding space, the algebra of the relation embedding space, and the interaction between relation and entity embeddings. Neglecting the geometry of relation embedding limits the optimization of entity and relation distribution leading to suboptimal performance of knowledge graph completion. To address this issue, we propose a new perspective in the design of KGEs by looking into the geometry of relation embedding space. The proposed method and its variants are developed on top of an existing framework, RotatE, from which we leverage the geometry of the relation embeddings by mutating the unit circle to an ellipse, and further generalize it with the concept of a butterfly curve, consecutively. Besides the theoretical abilities of the model in preserving topological and relational patterns, the experiments on the WN18RR, FB15K-237 and YouTube benchmarks showed that this new family of KGEs can challenge or outperform state-of-the-art models.