Process Reward Models (PRMs) aim to identify and mitigate intermediate errors in the reasoning processes in mathematical reasoning of Large Language Models (LLMs).However, the development of effective PRMs faces significant challenges, particularly in data annotation and evaluation methodologies.In this paper, through extensive experiments, we demonstrate that commonly used Monte Carlo (MC) estimation-based data synthesis for PRMs typically yields inferior performance and generalization compared to LLM-as-a-judge and human annotation methods.Furthermore, we identify potential biases in conventional Best-of-N (BoN) evaluation strategies for PRMs.To address these challenges, we develop a consensus filtering mechanism that effectively integrates MC estimation with LLM-as-a-judge and advocates a more comprehensive evaluation framework that combines response-level and step-level metrics. Based on the mechanisms, we significantly improve both model performance and data efficiency in the BoN evaluation and the step-wise error identification task.Finally, we release a new state-of-the-art PRM that outperforms existing open-source alternatives and provides practical guidelines for future research.
论文检索
输入标题、作者或关键词，从 3,086 篇学术成果中精准定位
ESF: Efficient Sensitive Fingerprinting for Black-Box Tamper Detection of Large Language Models
PDF ↗The rapid adoption of large language models (LLMs) in diverse applications has intensified concerns over their security and integrity, especially in cloud environments where internal model parameters are inaccessible to users. Traditional tamper detection methods, designed for deterministic classification models, fail to address the output randomness and massive parameter spaces characteristic of LLMs. In this paper, we introduce Efficient Sensitive Fingerprinting (ESF), the first fingerprinting method tailored for black-box tamper detection of LLMs. ESF generates fingerprint samples by optimizing output sensitivity at selected detection token positions and leverages Randomness-Set Consistency Checking (RSCC) to accommodate inherent output randomness. Furthermore, a novel Max Coverage Strategy (MCS) is proposed to select an optimal set of fingerprint samples that maximizes joint sensitivity to tampering. Grounded in a rigorous theoretical framework, ESF is both computationally efficient and scalable to large models. Extensive experiments across state-of-the-art LLMs demonstrate that ESF reliably detects tampering, such as fine-tuning, model compression, and backdoor injection, with a detection rate exceeding 99.2% using 5 fingerprint samples, thereby offering a robust solution for securing cloud-based AI systems.
DAGS: A Dependency-Based Dual-Attention and Global Semantic Improvement Framework for Metaphor Recognition
PDF ↗Current metaphor recognition mainly rely on Metaphor Detection Theory (MDT), such as the Metaphor Identification Procedure, which recognizes metaphors by comparing the basic meaning of target word with context meaning. Existing studies have gradually adopted literal annotations to model basic meanings, rejecting the aggregated meanings of target words. However, these methods ignore the problem of interference caused by literal annotations, and do not make full use of semantic expression relations of MDT, making the models difficult to detect and generalize. To address these challenges, we propose a dependency-based Dual-Attention and Global Semantic Improvement (DAGS) framework. DAGS first extracts literal annotations of target words as basic meaning from several mainstream corpora. Then, we apply dependency tree and dual-attention while filtering on input sentences and basic meanings. Finally, we improve the MDT to further consider the global semantic relationship on contexts. The DAGS can not only extract features from multiple information sources but alsoeffectively removes redundancy, while focusing on mission-critical information. We achieve state-of-the-art on several mainstream metaphor datasets (e.g., VUA ALL, VUAverb, TroFi and PSUCMC), which suggests that filtering and global semantic improvement of contexts is crucial for enhancing metaphor recognition performance.
Enhancing Transformation from Natural Language to Signal Temporal Logic Using LLMs with Diverse External Knowledge
PDF ↗Temporal Logic (TL), especially Signal Temporal Logic (STL), enables precise formal specification, making it widely used in cyber-physical systems such as autonomous driving and robotics. Automatically transforming NL into STL is an attractive approach to overcome the limitations of manual transformation, which is time-consuming and error-prone. However, due to the lack of datasets, automatic transformation currently faces significant challenges and has not been fully explored. In this paper, we propose a NL-STL dataset named STL-Diversity-Enhanced (STL-DivEn), comprising 16,000 samples enriched with diverse patterns. To develop the dataset, we first manually create a small-scale seed set of NL-STL pairs. Next, representative examples are identified through clustering and used to guide large language models (LLMs) in generating additional NL-STL pairs. Finally, diversity and accuracy are ensured through rigorous rule-based filters and human validation. Furthermore, we introduce the Knowledge-Guided STL Transformation (KGST) framework, a novel approach for transforming natural language into STL, involving a generate-then-refine process based on external knowledge. Statistical analysis shows that the STL-DivEn dataset exhibits more diversity than the existing NL-STL dataset. Moreover, both metric-based and human evaluations indicate that our KGST approach outperforms baseline models in transformation accuracy on STL-DivEn and DeepSTL datasets.
Neural retrieval methods using transformer-based pre-trained language models have advanced multilingual and cross-lingual retrieval. However, their effectiveness for low-resource, morphologically rich languages such as Amharic remains underexplored due to data scarcity and suboptimal tokenization. We address this gap by introducing Amharic-specific dense retrieval models based on pre-trained Amharic BERT and RoBERTa backbones. Our proposed RoBERTa-Base-Amharic-Embed model (110M parameters) achieves a 17.6% relative improvement in MRR@10 and a 9.86% gain in Recall@10 over the strongest multilingual baseline, Arctic Embed 2.0 (568M parameters). More compact variants, such as RoBERTa-Medium-Amharic-Embed (42M), remain competitive while being over 13\times smaller. Additionally, we train a ColBERT-based late interaction retrieval model that achieves the highest MRR@10 score (0.843) among all evaluated models. We benchmark our proposed models against both sparse and dense retrieval baselines to systematically assess retrieval effectiveness in Amharic. Our analysis highlights key challenges in low-resource settings and underscores the importance of language-specific adaptation. To foster future research in low-resource IR, we publicly release our dataset, codebase, and trained models at https://github.com/kidist-amde/amharic-ir-benchmarks.
Semantic Evaluation of Multilingual Data-to-Text Generation via NLI Fine-Tuning: Precision, Recall and F1 scores
PDF ↗Performance in the KG-to-Text task has improved over the years, particularly in English. However, models are still prone to mistakes like Additions and Omissions. Furthermore, few languages are taken into account since both train and test data are not readily available. In this paper, we hope to facilitate the development and improvement of multilingual KG-to-Text models by providing a multilingual evaluation framework that is reference-less (no need for test data) and permits estimating how much a KG-to-Text Model under- (omission) or over- (addition) generates. We focus on two high (English, Russian) and five low (Breton, Irish, Maltese, Welsh, Xhosa) resource languages and show that our metric has fair to moderate correlation with reference-based metrics, positioning it as a consistent alternative when no references are available. We also show that our metric outperforms prior reference-less metrics in correlation with existing human judgments. Additional human evaluation shows moderate to strong correlation with human annotators in assessing precision and recall at a higher granularity level than shown in previous studies. Since our metric provides scores for precision and recall, it helps better assess the level of over- or under-generation of multilingual KG-to-Text models.
Large Language Models (LLMs) have significantly impacted various domains, especially through organized LLM-driven autonomous agents. A representative scenario is in software development, where agents can collaborate in a team like humans, following predefined phases to complete sub-tasks sequentially. However, for an agent team, each phase yields only one possible outcome. This results in the completion of only one development chain, thereby losing the opportunity to explore multiple potential decision paths within the solution space. Consequently leading to suboptimal results or extensive trial and error. To address this, we introduce Cross-Team Orchestration (Croto), a scalable multi-team framework that enables orchestrated teams to jointly propose various task-oriented solutions and interact with their insights in a self-independence while cross-team collaboration environment for superior solutions generation. Experiments reveal a notable increase in software quality compared to state-of-the-art baselines. We further tested our framework on story generation tasks, which demonstrated a promising generalization ability of our framework in other domains. The code and data is available at https://github.com/OpenBMB/ChatDev/tree/macnet
Knowledge editing (KE) methods offer an efficient way to modify knowledge in large language models. Current KE evaluations typically assess editing success by considering only the edited knowledge without any preceding contexts. In real-world applications, however, preceding contexts often trigger the retrieval of the original knowledge and undermine the intended edit. To address this issue, we have developed CHED—a benchmark designed to evaluate the context robustness of KE methods. Evaluations on CHED show that they often fail when preceding contexts are present. To mitigate this shortcoming, we introduce CoRE, a KE method designed to strengthen context robustness by minimizing context-sensitive variance in hidden states of the model for edited knowledge. This method not only improves the editing success rate in situations where a preceding context is present but also preserves the overall capabilities of the model. We also provide an in-depth analysis of the differing impacts of preceding contexts when introduced as user utterances versus assistant responses, and we dissect attention-score patterns to assess how specific tokens influence editing success. We release our dataset and code at [https://github.com/holi-lab/CoRE](https://github.com/holi-lab/CoRE).
Large Language Models (LLMs) are transforming healthcare through LLM-based agents that can understand and assist with medical tasks. This survey examines the architectures, applications, and challenges of LLM-based agents in medicine. We analyze key components including system profiles, clinical planning, medical reasoning frameworks, and external capacity enhancement. The survey covers major applications in clinical decision support, medical documentation, training simulations, and healthcare service optimization, along with evaluation frameworks and metrics. While these agents show promise in enhancing healthcare delivery, challenges remain in hallucination management, multimodal integration, implementation, and ethics. We conclude by highlighting future directions in medical reasoning, physical system integration, and training simulations, providing researchers and practitioners with a structured overview of the field’s current state and prospects.
We introduce TableLLM, a robust large language model (LLM) with 8 billion parameters, purpose-built for proficiently handling tabular data manipulation tasks, whether they are embedded within documents or spreadsheets, catering to real-world office scenarios. We propose a distant supervision method for training, which comprises a reasoning process extension strategy, aiding in training LLMs to understand reasoning patterns more effectively as well as a cross-way validation strategy, ensuring the quality of the automatically generated data. To evaluate the performance of TableLLM, we have crafted benchmarks tailored to address both document and spreadsheet formats as well as constructed a well-organized evaluation pipeline capable of handling both scenarios. Thorough evaluations underscore the advantages of TableLLM when compared to various existing general-purpose and tabular data-focused LLMs. We have publicly released the model checkpoint, source code, benchmarks, and a web application for user interaction on this anonymized repository.
The application of role-playing large language models (LLMs) is rapidly expanding in both academic and commercial domains, driving an increasing demand for high-precision role-playing models. Simultaneously, the rapid advancement of reasoning techniques has continuously pushed the performance boundaries of LLMs. This intersection of practical role-playing demands and evolving reasoning capabilities raises an important research question: Can reasoning techniques enhance the role-playing capabilities of LLMs?” To address this, we conduct a comprehensive study using 6 role-playing benchmarks, 24 LLMs, and 3 distinct role-playing strategies, comparing the effectiveness of direct zero-shot role-playing, role-playing with Chain-of-Thought (CoT), and role-playing using reasoning-optimized LLMs. Our findings reveal that CoT may reduce role-playing performance, reasoning-optimized LLMs are unsuitable for role-playing, reasoning ability disrupts the role-playing scaling law, and large models still lack proficiency in advanced role-playing. Furthermore, based on extensive experimental results, we propose two promising future research directions: Role-aware Chain-of-Thought for improving role-playing LLMs and Reinforcement Learning for role-playing LLMs, aiming to enhance the adaptability, consistency, and effectiveness of role-playing LLMs for both research and real-world applications.
Reliable responses from large language models (LLMs) require adherence to user instructions and retrieved information. While alignment techniques help LLMs align with human intentions and values, improving context-faithfulness through alignment remains underexplored. To address this, we propose Context-DPO, the first alignment method specifically designed to enhance LLMs’ context-faithfulness. We introduce ConFiQA, a benchmark that simulates Retrieval-Augmented Generation (RAG) scenarios with knowledge conflicts to evaluate context-faithfulness. By leveraging faithful and stubborn responses to questions with provided context from ConFiQA, our Context-DPO aligns LLMs through direct preference optimization. Extensive experiments demonstrate that our Context-DPO significantly improves context-faithfulness, achieving 35% to 280% improvements on popular open-source models. Further analysis demonstrates that Context-DPO preserves LLMs’ generative capabilities while providing interpretable insights into context utilization.
LLM Unlearning plays a crucial role in removing sensitive information from language models to mitigate potential misuse. However, previous approaches often treat nonsensical responses or template-based refusals (e.g., “Sorry, I cannot answer.”) as the unlearning target, which can give the impression of deliberate information suppression, making the process even more vulnerable to attacks and jailbreaks. Moreover, most methods rely on auxiliary models or retaining datasets, which adds complexity to the unlearning process. To address these challenges, we propose MEOW, a streamlined and stealthy unlearning method that eliminates the need for auxiliary models or retaining data while avoiding leakage through its innovative use of inverted facts. These inverted facts are generated by an offline LLM and serve as fine-tuning labels. Meanwhile, we introduce MEMO, a novel metric that measures the model’s memorization, to select optimal fine-tuning targets. The use of inverted facts not only maintains the covert nature of the model but also ensures that sensitive information is effectively forgotten without revealing the target data. Evaluated on the ToFU Knowledge Unlearning dataset using Llama2-7B-Chat and Phi-1.5, MEOW outperforms baselines in forgetting quality while preserving model utility. MEOW also maintains strong performance across NLU and NLG tasks and demonstrates superior resilience to attacks, validated via the Min-K% membership inference method.
End-to-end Large Speech Language Models (**LSLMs**) demonstrate strong potential in response latency and speech comprehension capabilities, showcasing general intelligence across speech understanding tasks. However, the ability to follow speech instructions has not been fully realized due to the lack of datasets and heavily biased training tasks. Leveraging the rich ASR datasets, previous approaches have used Large Language Models (**LLMs**) to continue the linguistic information of speech to construct speech instruction datasets. Yet, due to the gap between LLM-generated results and real human responses, the continuation methods further amplify these shortcomings. Given the high costs of collecting and annotating speech instruction datasets by humans, using speech synthesis to construct large-scale speech instruction datasets has become a balanced and robust alternative. Although modern Text-To-Speech (**TTS**) models have achieved near-human-level synthesis quality, it is challenging to appropriately convert out-of-distribution text instruction to speech due to the limitations of the training data distribution in TTS models. To address this issue, we propose a query rewriting framework with multi-LLM knowledge fusion, employing multiple agents to annotate and validate the synthesized speech, making it possible to construct high-quality speech instruction datasets without relying on human annotation. Experiments show that this method can transform text instructions into distributions more suitable for TTS models for speech synthesis through zero-shot rewriting, increasing data usability from 72% to 93%. It also demonstrates unique advantages in rewriting tasks that require complex knowledge and context-related abilities.
The conceptual knowledge in Large Language Models (LLMs) can become outdated over time, and concept editing is often an option. Current evaluations on conceptual knowledge editing primarily focus on whether the definitions of concepts are successfully edited, neglecting the impact on the model’s related beliefs. To address this gap, we introduce a benchmark called RelEdit, which includes criteria and questions to assess both concept-level and instance-level relational reasoning abilities of edited models. Our findings reveal that existing knowledge editing methods struggle to reason about related conceptual knowledge effectively. Additionally, we introduce a simple memory-based in-context editing baseline, MICE, which prompts the language model to generate answers that align with the stored edited concepts in external memory. In addition, we find that MICE obtains the best scores on our benchmark, suggesting a promising research direction for model editing.
STATE ToxiCN: A Benchmark for Span-level Target-Aware Toxicity Extraction in Chinese Hate Speech Detection
PDF ↗The proliferation of hate speech has caused significant harm to society. The intensity and directionality of hate are closely tied to the target and argument it is associated with. However, research on hate speech detection in Chinese has lagged behind, and existing datasets lack span-level fine-grained annotations. Furthermore, the lack of research on Chinese hateful slang poses a significant challenge. In this paper, we provide two valuable fine-grained Chinese hate speech detection research resources. First, we construct a Span-level Target-Aware Toxicity Extraction dataset (STATE ToxiCN), which is the first span-level Chinese hate speech dataset. Secondly, we evaluate the span-level hate speech detection performance of existing models using STATE ToxiCN. Finally, we conduct the first study on Chinese hateful slang and evaluate the ability of LLMs to understand hate semantics. Our work contributes valuable resources and insights to advance span-level hate speech detection in Chinese.
Interactive Fiction games (IF games) are where players interact through natural language commands. While recent advances in Artificial Intelligence agents have reignited interest in IF games as a domain for studying decision-making, existing approaches prioritize task-specific performance metrics over human-like comprehension of narrative context and gameplay logic. This work presents a cognitively inspired framework that guides Large Language Models (LLMs) to learn and play IF games systematically. Our proposed **L**earning to **P**lay **L**ike **H**umans (LPLH) framework integrates three key components: (1) structured map building to capture spatial and narrative relationships, (2) action learning to identify context-appropriate commands, and (3) feedback-driven experience analysis to refine decision-making over time. By aligning LLMs-based agents’ behavior with narrative intent and commonsense constraints, LPLH moves beyond purely exploratory strategies to deliver more interpretable, human-like performance. Crucially, this approach draws on cognitive science principles to more closely simulate how human players read, interpret, and respond within narrative worlds. As a result, LPLH reframes the IF games challenge as a learning problem for LLMs-based agents, offering a new path toward robust, context-aware gameplay in complex text-based environments.
As Large Language Models are increasingly deployed in high-stakes domains, their ability to detect false assumptions and reason critically is crucial for ensuring reliable outputs. False-premise questions (FPQs) serve as an important evaluation method by exposing cases where flawed assumptions lead to incorrect responses. While existing benchmarks focus on single-hop FPQs, real-world reasoning often requires multi-hop inference, where models must verify consistency across multiple reasoning steps rather than relying on surface-level cues. To address this gap, we introduce MultiHoax, a benchmark for evaluating LLMs’ ability to handle false premises in complex, multi-step reasoning tasks. Our dataset spans seven countries and ten diverse knowledge categories, using Wikipedia as the primary knowledge source to enable cross-regional factual reasoning. Experiments reveal that state-of-the-art LLMs struggle to detect false premises across different countries, knowledge categories, and multi-hop reasoning types, highlighting the need for improved false premise detection and more robust multi-hop reasoning capabilities in LLMs.
Honest alignment refers to the ability of a language model to truthfully convey its knowledge limitations by appropriately refusing to answer questions when it lacks sufficient information. Existing solutions, such as prompt engineering and fine-tuning, face limitations: the former provides only marginal improvements, while the latter struggles to enhance honesty when annotated data is scarce.To overcome the above limitations, we propose , a novel framework that enhances honesty through weak-to-strong generalization. Specifically, we train the strong LLMs under weak model supervision to improve their honesty. For the weak model, we employ a learning-to-rank strategy to train a “honest head”, which learns to select the most honest response among model’s outputs generated through beam search. For the strong LLM, we leverage the self-labeled dataset to update its parameters. Our proposal requires only minimal training data to train the weak honest model, yet achieve decent performance for labeling data. In addition, it enables the strong LLMs to have the capabilities to generalize even facing with the flawed label data. Extensive experiments show significantly boosts honest alignment in large models even with limited labeled data. Our code is available at https://github.com/zewanfaan/WHAT_Honesty.
DI-BENCH: Benchmarking Large Language Models on Dependency Inference with Testable Repositories at Scale
PDF ↗Large Language Models have advanced automated software development, however, it remains a challenge to correctly infer dependencies, namely, identifying the internal components and external packages required for a repository to successfully run. Existing studies highlight that dependency-related issues cause over 40% of observed runtime errors on the generated repository. To address this, we introduce DI-BENCH, a large-scale benchmark and evaluation framework specifically designed to assess LLMs’ capability on dependency inference. The benchmark features 581 repositories with testing environments across Python, C#, Rust, and JavaScript. Extensive experiments with textual and execution-based metrics reveal that the current best-performing model achieves only a 48% execution pass rate on Python, indicating significant room for improvement. DI-BENCH establishes a new viewpoint for evaluating LLM performance on repositories, paving the way for more robust end-to-end software synthesis.