Large language models (LLMs) show strong performance in medical question answering, but their computational costs limit deployment under resource constraints. Therefore, an increasing number of studies focus on distilling knowledge from LLMs into smaller models to improve performance in medical question answering. However, existing methods often fail to identify truly useful knowledge to distill, making it difficult to focus distillation on the student’s real weaknesses. As a result, much of the distillation budget is spent on already acquired knowledge rather than on weaknesses that still limit performance. To address this problem, we propose LEAD, a Learnable Error-Aware Distillation framework that precisely improves the weaknesses of student models. First, LEAD identifies questions answered incorrectly by the student model to determine which weaknesses require teacher guidance. Then, LEAD generates teacher rationales and similar samples to construct candidate supervision for these errors. Through supervision refinement, reliable and learnable training samples are retained from the candidate supervision. Finally, LEAD trains the student on a balanced sample mixture to correct learnable errors while preserving previously acquired knowledge. Experiments on four medical question answering benchmarks show that LEAD outperforms existing distillation methods.

Large Language Models often exhibit dishonest behaviors–confidently hallucinating answers to questions beyond their knowledge boundary or hedging on queries they have mastered. Although existing mitigation methods have progressed to establishing the model knowledge boundary, this boundary is still typically treated as a coarse “known vs. unknown” distinction, leaving a finer-grained question underexplored: Does a failure reflect missing knowledge, or merely under-expressed capability? In this paper, we propose KCAPO, a framework that aligns the model’s output behavior with its latent capabilities. It estimates query-level solvability with a consistency-supervised hidden-state probe, then uses its mismatch with response confidence and factual behavior to refine the knowledge boundary into four capability–behavior cases. Based on this fine-grained characterization, we construct preference data to dynamically "Wake Up" dormant knowledge or "Suppress" blind guessing. Furthermore, we introduce an adaptive margin into the DPO objective, which prioritizes correcting responses with severe cognitive dissonance. Extensive experiments on short-form QA and refusal benchmarks show that KCAPO reduces hallucinations without compromising utility and improves selective abstention.

Retrieval-Augmented Generation (RAG) mitigates hallucinations in large language models by incorporating external knowledge. However, retrieval does not always return relevant documents and may return noisy ones. Indiscriminately retrieving and utilizing this external knowledge can interfere with the model’s originally correct reasoning. In this work, we propose Dual-Decision Retrieval-Augmented Generation (D²-RAG), which integrates multi-dimensional uncertainty estimation to decide whether to retrieve and employs adaptive contrastive decoding to handle retrieved contexts of varying quality. Specifically, we first integrate uncertainty estimation scores that assess model uncertainty from multiple perspectives, construct them into a comprehensive feature vector, and train a lightweight retrieval decision model to accurately identify the model’s knowledge boundaries and determine whether to retrieve. Subsequently, we dynamically adjust the contrastive decoding strategy based on the utility of retrieved contexts to enhance the utilization of relevant contexts while suppressing interference from noisy contexts. Extensive experiments on four medical question-answering datasets demonstrate that D²-RAG significantly outperforms baselines, enabling retrieval-augmented Llama3.1-8B to surpass non-retrieval-augmented Llama3.1-70B on the MedMCQA dataset. The source code is available on https://github.com/zakelawen/d--rag.

Large language models (LLMs) have demonstrated strong reasoning abilities, but hallucinations and unverifiable intermediate reasoning steps remain key challenges. In knowledge graph question answering (KGQA), entity-relation paths from knowledge graphs (KGs) provide structured and verifiable evidence for grounded reasoning. However, existing KG-enhanced LLM methods commonly linearize all retrieved reasoning paths into textual prompts, introducing noise and underexploiting their ordered structure. To address these limitations, we propose PathSoft, a hybrid soft prompting framework that combines textual path evidence with soft prompts. PathSoft first employs PathMLP to select reliable reasoning paths, and then uses PathSoftAdapter to convert the selected paths into continuous soft prompts while preserving their ordered path structure. Finally, the resulting soft prompt tokens are combined with the textual prompt at the embedding layer of the frozen LLM, enabling it to exploit both readable path evidence and path structure information. Experiments on WebQSP and CWQ show that PathSoft outperforms existing KGQA baselines. The code is available at https://github.com/tengyan1786-blip/PathSoft.

With the generative capabilities of large language models (LLMs) reshaping the information ecosystem, the concern with the sociological validity of claim detection benchmarks is increasing. Current claim detection benchmarks predominantly treat claims as static textual artifacts, overlooking the sociological etiology of how information naturally emerges and mutates. In this paper, we propose an evolutionary paradigm that models claims as socially evolving entities. In specific, we introduce a socially generative framework for synthetic claim generation, a multi-agent simulation grounded in the Open Claims Model. By decomposing claims into context, utterance, and proposition, our approach enables the precise simulation of unmitigated propagation to capture truth decay, and intervened propagation with multi-auditor oversight for targeted generation. Furthermore, we propose the background-user-perspective (BUP) framework, which reformulates check-worthiness as a condition-dependent probability rooted in social environment. Experiments on our datasets verify the data quality and reveal how network topology and user attributes systematically shape veracity drift.

The alignment of large language models (LLMs) aims to ensure their outputs adhere to human values, ethical standards, and legal norms. Traditional alignment methods often rely on resource-intensive fine-tuning (FT), which may suffer from knowledge degradation and face challenges in scenarios where the model accessibility or computational resources are constrained. In contrast, training-free (TF) alignment techniques—leveraging in-context learning, decoding-time adjustments, and post-generation corrections—offer a promising alternative by enabling alignment without heavily retraining LLMs, making them adaptable to both open-source and closed-source environments. This paper presents the first systematic review of TF alignment methods, categorizing them by stages of pre-decoding, in-decoding, and post-decoding. For each stage, we provide a detailed examination from the viewpoint of LLMs and multimodal LLMs (MLLMs), highlighting their mechanisms and limitations. Furthermore, we identify key challenges and future directions, paving the way for more inclusive and effective TF alignment techniques. By synthesizing and organizing the rapidly growing body of research, this survey offers a guidance for practitioners and advances the development of safer and more reliable LLMs.

Argumentative essay generation (AEG) is a complex task that requires advanced semantic understanding, logical reasoning, and organized integration of perspectives. Despite showing a promising performance, current efforts often overlook the dynamical and hierarchical nature of structural argumentative planning, and struggle with flexible rhetorical expression, leading to limited argument divergence and rhetorical optimization. Inspired by human debate behavior and Bitzer’s rhetorical situation theory, we propose a debate-driven rhetorical framework for argumentative writing. The uniqueness lies in three aspects: (1) dynamic assesses the divergence of viewpoints and progressively reveals the hierarchical outline of arguments based on a depth-then-breadth paradigm, improving the perspective divergence within argumentation; (2) simulates human debate through iterative defender-attacker interactions, improving the logical coherence of arguments; (3) incorporates Bitzer’s rhetorical situation theory to flexibly select appropriate rhetorical techniques, enabling the rhetorical expression. Experiments on four benchmarks validate that our approach significantly improves logical depth, argumentative diversity, and rhetorical persuasiveness over existing state-of-the-art models.