| Total: 3490
Although LLM-based conversational agents demonstrate strong fluency and coherence, they still produce undesirable behaviors (errors) that are challenging to prevent from reaching users during deployment. Recent research leverages large language models (LLMs) to detect errors and guide response-generation models toward improvement. However, current LLMs struggle to identify errors not explicitly specified in their instructions, such as those arising from updates to the response-generation model or shifts in user behavior. In this work, we introduce Automated Error Discovery, a framework for detecting and defining errors in conversational AI, and propose SEEED (Soft Clustering Extended Encoder-Based Error Detection), as an encoder-based approach to its implementation. We enhance the Soft Nearest Neighbor Loss by amplifying distance weighting for negative samples and introduce Label-Based Sample Ranking to select highly contrastive examples for better representation learning. SEEED outperforms adapted baselines—including GPT-4o and Phi-4—across multiple error-annotated dialogue datasets, improving the accuracy for detecting unknown errors by up to 8 points and demonstrating strong generalization to unknown intent detection.
A large number of studies rely on closed-style multiple-choice surveys to evaluate cultural alignment in Large Language Models (LLMs). In this work, we challenge this constrained evaluation paradigm and explore more realistic, unconstrained approaches. Using the World Values Survey (WVS) and Hofstede Cultural Dimensions as case studies, we demonstrate that LLMs exhibit stronger cultural alignment in less constrained settings, where responses are not forced. Additionally, we show that even minor changes, such as reordering survey choices, lead to inconsistent outputs, exposing the limitations of closed-style evaluations. Our findings advocate for more robust and flexible evaluation frameworks that focus on specific cultural proxies, encouraging more nuanced and accurate assessments of cultural alignment in LLMs.
Stories play a pivotal role in human communication, shaping beliefs and morals, particularly in children. As parents increasingly rely on large language models (LLMs) to craft bedtime stories, the presence of cultural and gender stereotypes in these narratives raises significant concerns. To address this issue, we present Biased Tales, a comprehensive dataset designed to analyze how biases influence protagonists’ attributes and story elements in LLM-generated stories. Our analysis uncovers striking disparities. When the protagonist is described as a girl (as compared to a boy), appearance-related attributes increase by 55.26%. Stories featuring non-Western children disproportionately emphasize cultural heritage, tradition, and family themes far more than those for Western children. Our findings highlight the role of sociocultural bias in making creative AI use more equitable and diverse.
Workload traces are essential to understand complex computer systems’ behavior and manage processing and memory resources. Since real-world traces are hard to obtain, synthetic trace generation is a promising alternative. This paper proposes a first-of-a-kind approach that relies on training a large language model (LLM) to generate synthetic workload traces, specifically microservice call graphs. To capture complex and arbitrary hierarchical structures and implicit constraints in such traces, we propose to train LLMs to generate recursively, making call graph generation a sequence of more manageable steps. To further enforce learning constraints on the traces and generate uncommon situations, we apply additional instruction tuning steps to align our model with the desired trace features. With this method, we train TraceLLM, an LLM for microservice trace generation, and demonstrate that it produces diverse, realistic traces under varied conditions, outperforming existing approaches in both accuracy and validity. The synthetically generated traces can effectively replace real data to optimize important microservice management tasks. Additionally, TraceLLM adapts to downstream trace-related tasks, such as predicting key trace features and infilling missing data.
Simplifying text while preserving its meaning is a complex yet essential task, especially in sensitive domain applications like legal texts. When applied to a specialized field, like the legal domain, preservation differs significantly from its role in regular texts. This paper introduces FrJUDGE, a new dataset to assess legal meaning preservation between two legal texts. It also introduces JUDGEBERT, a novel evaluation metric designed to assess legal meaning preservation in French legal text simplification. JUDGEBERT demonstrates a superior correlation with human judgment compared to existing metrics. It also passes two crucial sanity checks, while other metrics did not: For two identical sentences, it always returns a score of 100%; on the other hand, it returns 0% for two unrelated sentences. Our findings highlight its potential to transform legal NLP applications, ensuring accuracy and accessibility for text simplification for legal practitioners and lay users.
Large and Transformer-based language models perform outstandingly in various downstream tasks. However, there is limited understanding regarding how these models internalize linguistic knowledge, so various linguistic benchmarks have recently been proposed to facilitate syntactic evaluation of language models across languages. This paper introduces QFrCoLA (Quebec-French Corpus of Linguistic Acceptability Judgments), a normative binary acceptability judgments dataset comprising 25,153 in-domain and 2,675 out-of-domain sentences. Our study leverages the QFrCoLA dataset and seven other linguistic binary acceptability judgment corpora to benchmark seven language models. The results demonstrate that, on average, fine-tuned Transformer-based LM are strong baselines for most languages and that zero-shot binary classification large language models perform poorly on the task. However, for the QFrCoLA benchmark, on average, a fine-tuned Transformer-based LM outperformed other methods tested. It also shows that pre-trained cross-lingual LLMs selected for our experimentation do not seem to have acquired linguistic judgment capabilities during their pre-training for Quebec French. Finally, our experiment results on QFrCoLA show that our dataset, built from examples that illustrate linguistic norms rather than speakers’ feelings, is similar to linguistic acceptability judgment; it is a challenging dataset that can benchmark LM on their linguistic judgment capabilities.
The value orientation of Large Language Models (LLMs) has been extensively studied, as it can shape user experiences across demographic groups.However, two key challenges remain: (1) the lack of systematic comparison across value probing strategies, despite the Multiple Choice Question (MCQ) setting being vulnerable to perturbations, and (2) the uncertainty over whether probed values capture in-context information or predict models’ real-world actions.In this paper, we systematically compare three widely used value probing methods: token likelihood, sequence perplexity, and text generation.Our results show that all three methods exhibit large variances under non-semantic perturbations in prompts and option formats, with sequence perplexity being the most robust overall.We further introduce two tasks to assess expressiveness: demographic prompting, testing whether probed values adapt to cultural context; and value–action agreement, testing the alignment of probed values with value-based actions.We find that demographic context has little effect on the text generation method, and probed values only weakly correlate with action preferences across all methods.Our work highlights the instability and the limited expressive power of current value probing methods, calling for more reliable LLM value representations.
Reinforcement learning from human feedback (RLHF) and, at its core, reward modeling have become a crucial part of training powerful large language models (LLMs). One commonly overlooked factor in training high-quality reward models (RMs) is the effect of the base model, which is becoming more challenging to choose given the rapidly growing pool of LLMs. In this work, we present a systematic analysis of the effect of base model selection on reward modeling performance. Our results show that the performance can be improved by up to 14% compared to the most common (i.e., default) choice. Moreover, we showcase the strong statistical relation between some existing benchmarks and downstream performances. We also demonstrate that the results from a small set of benchmarks could be combined to boost the model selection (+18% on average in the top 5-10). Lastly, we illustrate the impact of different post-training steps on the final performance and explore using estimated data distributions to reduce performance prediction error.
When solving NLP tasks with limited labelled data, researchers typically either use a general large language model without further update, or use a small number of labelled samples to tune a specialised smaller model. In this work, we answer an important question – how many labelled samples are required for the specialised small models to outperform general large models, while taking the performance variance into consideration. By observing the behaviour of fine-tuning, instruction-tuning, prompting and in-context learning on 8 language models, we identify such performance break-even points across 8 representative text classification tasks of varying characteristics. We show that the specialised models often need only few samples (on average 100) to be on par or better than the general ones. At the same time, the number of required labels strongly depends on the dataset or task characteristics, with fine-tuning on binary datasets requiring significantly more samples. When performance variance is taken into consideration, the number of required labels increases on average by 100 - 200%. Finally, larger models do not consistently lead to better performance and lower variance, with 4-bit quantisation having negligible impact.
Determining faithfulness of a claim to a source document is an important problem across many domains. This task is generally treated as a binary judgment of whether the claim is supported or unsupported in relation to the source. In many cases, though, whether a claim is supported can be ambiguous. For instance, it may depend on making inferences from given evidence, and different people can reasonably interpret the claim as either supported or unsupported based on their agreement with those inferences. Forcing binary labels upon such claims lowers the reliability of evaluation. In this work, we reframe the task to manage the subjectivity involved with factuality judgments of ambiguous claims. We introduce LLM-generated edits of summaries as a method of providing a nuanced evaluation of claims: how much does a summary need to be edited to be unambiguous? Whether a claim gets rewritten and how much it changes can be used as an automatic evaluation metric, the Ambiguity Rewrite Metric (ARM), with a much richer feedback signal than a binary judgment of faithfulness. We focus on the area of narrative summarization as it is particularly rife with ambiguity and subjective interpretation. We show that ARM produces a 21% absolute improvement in annotator agreement on claim faithfulness, indicating that subjectivity is reduced.
Evaluating the pedagogical capabilities of AI-based tutoring models is critical for making guided progress in the field. Yet, we lack a reliable, easy-to-use, and simple-to-run evaluation that reflects the pedagogical abilities of models. To fill this gap, we present MathTutorBench, an open-source benchmark for holistic tutoring model evaluation. MathTutorBench contains a collection of datasets and metrics that broadly cover tutor abilities as defined by learning sciences research in dialog-based teaching. To score the pedagogical quality of open-ended teacher responses, we train a reward model and show it can discriminate expert from novice teacher responses with high accuracy. We evaluate a wide set of closed- and open-weight models on MathTutorBench and find that subject expertise, indicated by solving ability, does not immediately translate to good teaching. Rather, pedagogy and subject expertise appear to form a trade-off that is navigated by the degree of tutoring specialization of the model. Furthermore, tutoring appears to become more challenging in longer dialogs, where simpler questioning strategies begin to fail. We release the benchmark, code, and leaderboard openly to enable rapid benchmarking of future models.
Deploying LLM-based agents in real-life applications often faces a critical challenge: the misalignment between agents’ behavior and user intent. Such misalignment may lead agents to unintentionally execute some critical actions that carry negative outcomes (e.g., accidentally triggering a buy-now in web shopping), resulting in undesirable or even irreversible consequences. Although addressing these issues is crucial, the preemptive detection and correction of misaligned actions remains relatively underexplored. To fill this gap, we introduce InferAct, a novel approach that leverages the belief reasoning ability of LLMs, grounded in Theory-of-Mind, to detect misaligned actions. Once the misalignment is detected, InferAct alerts users for timely correction, preventing adverse outcomes and enhancing the reliability of LLM agents’ decision-making processes. Experiments on three widely used tasks demonstrate InferAct achieves up to 20% improvements on Marco-F1 against baselines in misaligned action detection. An in-depth evaluation of misalignment correction further highlights InferAct‘s effectiveness in improving agent alignment.
LLMs increasingly engage with psychological instruments, yet how they represent constructs internally remains poorly understood. We introduce a novel approach to “fingerprinting” LLMs through their factor correlation patterns on standardized psychological assessments to deepen the understanding of LLMs constructs representation. Using the Humor Style Questionnaire as a case study, we analyze how six LLMs represent and correlate humor-related constructs to survey participants. Our results show that they exhibit little similarity to human response patterns. In contrast, participants’ subsamples demonstrate remarkably high internal consistency. Exploratory graph analysis further confirms that no LLM successfully recovers the four constructs of the Humor Style Questionnaire. These findings suggest that despite advances in natural language capabilities, current LLMs represent psychological constructs in fundamentally different ways than humans, questioning the validity of application as human simulacra.
Although Contrastive Language-Image Pre-training (CLIP) exhibits strong performance across diverse vision tasks, its application to person representation learning faces two critical challenges: (i) the scarcity of large-scale annotated vision-language data focused on person-centric images, and (ii) the inherent limitations of global contrastive learning, which struggles to maintain discriminative local features crucial for fine-grained matching while remaining vulnerable to noisy text tokens. This work advances CLIP for person representation learning through synergistic improvements in data curation and model architecture. First, we develop a noise-resistant data construction pipeline that leverages the in-context learning capabilities of MLLMs to automatically filter and caption web-sourced images. This yields WebPerson, a large-scale dataset of 5M high-quality person-centric image-text pairs. Second, we introduce the GA-DMS (Gradient-Attention Guided Dual-Masking Synergetic) framework, which improves cross-modal alignment by adaptively masking noisy textual tokens based on the gradient-attention similarity score. Additionally, we incorporate masked token prediction objectives that compel the model to predict informative text tokens, enhancing fine-grained semantic representation learning. Extensive experiments show that GA-DMS achieves state-of-the-art performance across multiple benchmarks. The data and pre-trained models are released at https://github.com/Multimodal-Representation-Learning-MRL/GA-DMS.
Large language models (LLMs) can transform education, but their optimization for direct question-answering often undermines effective pedagogy which requires strategically withholding answers. To mitigate this, we propose an online reinforcement learning (RL)-based alignment framework that can quickly adapt LLMs into effective tutors using simulated student-tutor interactions by emphasizing pedagogical quality and guided problem-solving over simply giving away answers. We use our method to train a 7B parameter tutor model without human annotations which reaches similar performance to larger proprietary models like LearnLM. We introduce a controllable reward weighting to balance pedagogical support and student solving accuracy, allowing us to trace the Pareto frontier between these two objectives. Our models better preserve reasoning capabilities than single-turn SFT baselines and can optionally enhance interpretability through thinking tags that expose the model’s instructional planning.
Knowledge Base Question Answering (KBQA) aims to extract accurate answers from the Knowledge Base (KB). Traditional Semantic Parsing (SP)-based methods are widely used but struggle with complex queries. Recently, large language models (LLMs) have shown promise in improving KBQA performance. However, the challenge of generating error-free logical forms remains, as skeleton, topic Entity, and relation Errors still frequently occur. To address these challenges, we propose CompKBQA(Component-wise Task Decomposition for Knowledge Base Question Answering), a novel framework that optimizes the process of fine-tuning a LLM for generating logical forms by enabling the LLM to progressively learn relevant sub-tasks like skeleton generation, topic entity generation, and relevant relations generation. Additionally, we propose R3, which retrieves and incorporates KB information into the process of logical form generation. Experimental evaluations on two benchmark KBQA datasets, WebQSP and CWQ, demonstrate that CompKBQA achieves state-of-the-art performance, highlighting the importance of task decomposition and KB-aware learning.
Aligning Large Language Models (LLMs) with human preferences is crucial in ensuring desirable and controllable model behaviors. Current methods, such as Reinforcement Learning from Human Feedback (RLHF) and Direct Preference Optimization (DPO), rely on the Bradley-Terry (B-T) model to maximize the likelihood of pairwise choices. However, when multiple responses are available, the B-T model fails to guarantee an accurate list ranking of the responses. To address this issue, we propose Permutative Preference Alignment (PPA), a novel offline listwise approach that incorporates the Normalized Discounted Cumulative Gain (NDCG)—a widely-used ranking metric—as an alternative training objective for LLM alignment. We develop an end-to-end alignment algorithm by approximating NDCG with a differentiable surrogate loss. Experiments demonstrate that PPA outperforms existing pairwise and listwise methods on evaluation sets and general benchmarks such as AlpacaEval. Furthermore, we show that NDCG-based approaches improve ranking accuracy more effectively than B-T-based methods and provide a theoretical explanation for this improvement.
Lyrics generation has garnered increasing attention within the artificial intelligence community. Our task focuses on generating harmonious Cantonese lyrics. Unlike other languages, Cantonese has a unique system of nine contours and six tones, making it essential to satisfy the harmony rules that ensure the alignment between the melody and the tonal contours of the lyrics when composing lyrics. Current research has not yet addressed the challenge of generating lyrics that adhere to Cantonese harmony rules. To tackle this issue, we propose ToneCraft, a novel framework for generating Cantonese lyrics that ensures tonal and melodic harmony. It enables LLMs to generate lyrics with a fixed character count while aligning with tonal and melodic structures. We present an algorithm that combines character-level control, melodic guidance, and a task-specific loss to achieve tonal harmony without compromising generation flexibility and quality. By incorporating domain-specific expertise, we leverage pure lyric datasets to train our model, eliminating the need for aligned data. Both objective evaluations and subjective assessments show that our generated lyrics align with melodic contours significantly better than existing methods. All code and data are available at: https://github.com/purepasser-by/ToneCraft.
We introduce SensorLLM, a two-stage framework that enables Large Language Models (LLMs) to perform human activity recognition (HAR) from sensor time-series data. Despite their strong reasoning and generalization capabilities, LLMs remain underutilized for motion sensor data due to the lack of semantic context in time-series, computational constraints, and challenges in processing numerical inputs. SensorLLM addresses these limitations through a Sensor-Language Alignment stage, where the model aligns sensor inputs with trend descriptions. Special tokens are introduced to mark channel boundaries. This alignment enables LLMs to capture numerical variations, channel-specific features, and data of varying durations, without requiring human annotations. In the subsequent Task-Aware Tuning stage, we refine the model for HAR classification, achieving performance that matches or surpasses state-of-the-art methods. Our results demonstrate that SensorLLM evolves into an effective sensor learner, reasoner, and classifier through human-intuitive Sensor-Language Alignment, generalizing across diverse HAR datasets. We believe this work establishes a foundation for future research on time-series and text alignment, paving the way for foundation models in sensor data analysis. Our codes are available at https://github.com/zechenli03/SensorLLM.
Continually updating model-based indexes in generative retrieval with new documents remains challenging, as full retraining is computationally expensive and impractical under resource constraints. We propose MixLoRA-DSI, a novel framework that combines an expandable mixture of Low-Rank Adaptation experts with a layer-wise out-of-distribution (OOD)-driven expansion strategy. Instead of allocating new experts for each new corpus, our proposed expansion strategy enables sublinear parameter growth by selectively introducing new experts only when significant number of OOD documents are detected. Experiments on NQ320k and MS MARCO Passage demonstrate that MixLoRA-DSI outperforms full-model update baselines, with minimal parameter overhead and substantially lower training costs.
The growing influence of video content as a medium for communication and misinformation underscores the urgent need for effective tools to analyze claims in multilingual and multi-topic settings. Existing efforts in misinformation detection largely focus on written text, leaving a significant gap in addressing the complexity of spoken text in video transcripts. We introduce ViClaim, a dataset of 1,798 annotated video transcripts across three languages (English, German, Spanish) and six topics. Each sentence in the transcripts is labeled with three claim-related categories: fact-check-worthy, fact-non-check-worthy, or opinion. We developed a custom annotation tool to facilitate the highly complex annotation process. Experiments with state-of-the-art multilingual language models demonstrate strong performance in cross-validation (macro F1 up to 0.896) but reveal challenges in generalization to unseen topics, particularly for distinct domains. Our findings highlight the complexity of claim detection in video transcripts. ViClaim offers a robust foundation for advancing misinformation detection in video-based communication, addressing a critical gap in multimodal analysis.
Large Language Models (LLMs) with web search capabilities show significant potential for deep research, yet current methods—brittle prompt engineering or RAG-based reinforcement learning in controlled environments—fail to capture real-world complexities. In this paper, we introduce DeepResearcher, the first comprehensive framework for end-to-end training of LLM-based deep research agents through scaling reinforcement learning (RL) in real-world environments with authentic web search interactions. Unlike RAG approaches reliant on fixed corpora, DeepResearcher trains agents to navigate the noisy, dynamic open web. We implement a specialized multi-agent architecture where browsing agents extract relevant information from various webpage structures and overcoming significant technical challenges. Extensive experiments on open-domain research tasks demonstrate that DeepResearcher achieves substantial improvements of up to 28.9 points over prompt engineering-based baselines and up to 7.2 points over RAG-based RL agents. Our qualitative analysis reveals emergent cognitive behaviors from end-to-end RL training, such as planning, cross-validation, self-reflection for research redirection, and maintain honesty when unable to find definitive answers. Our results highlight that end-to-end training in real-world web environments is fundamental for developing robust research capabilities aligned with real-world applications. The source codefor DeepResearcher is released at: https://github.com/GAIR-NLP/DeepResearcher.
Knowledge Graph (KG) reasoning, which aims to infer new facts from structured knowledge repositories, plays a vital role in Natural Language Processing (NLP) systems. Its effectiveness critically depends on constructing informative and contextually relevant reasoning paths. However, existing graph neural networks (GNNs) often adopt rigid, query-agnostic path-exploration strategies, limiting their ability to adapt to diverse linguistic contexts and semantic nuances. To address these limitations, we propose MoKGR, a mixture-of-experts framework that personalizes path exploration through two complementary components: (1) a mixture of length experts that adaptively selects and weights candidate path lengths according to query complexity, providing query-specific reasoning depth; and (2) a mixture of pruning experts that evaluates candidate paths from a complementary perspective, retaining the most informative paths for each query. Through comprehensive experiments on diverse benchmark, MoKGR demonstrates superior performance in both transductive and inductive settings, validating the effectiveness of personalized path exploration in KGs reasoning.
Stance detection, a critical task in Natural Language Processing (NLP), aims to identify the attitude expressed in text toward specific targets. Despite advancements in Large Language Models (LLMs), challenges such as limited interpretability and handling nuanced content persist. To address these issues, we propose the Multi-Path Reasoning Framework (MPRF), a novel framework that generates, evaluates, and integrates multiple reasoning paths to improve accuracy, robustness, and transparency in stance detection. Unlike prior work that relies on single-path reasoning or static explanations, MPRF introduces a structured end-to-end pipeline: it first generates diverse reasoning paths through predefined perspectives, then dynamically evaluates and optimizes each path using LLM-based scoring, and finally fuses the results via weighted aggregation to produce interpretable and reliable predictions. Extensive experiments on the SEM16, VAST, and PStance datasets demonstrate that MPRF outperforms existing models. Ablation studies further validate the critical role of MPRF’s components, highlighting its effectiveness in enhancing interpretability and handling complex stance detection tasks.
Large language models (LLMs) acquire substantial world knowledge during pre-training, which is further shaped by post-training techniques such as supervised fine-tuning (SFT). However, the impact of SFT on a model’s knowledge remains underexplored, limiting our ability to control knowledge behavior in fine-tuned models. To address this gap, we evaluate closed-book question answering (CBQA) performance across five LLMs from the LLaMA-2 and LLaMA-3 families. Surprisingly, models fine-tuned on 1,920 samples perform up to 14% worse than those fine-tuned on only 240 samples. Furthermore, varying the level of knowledge mastery in the fine-tuning data leads to performance fluctuations of over 12%. To investigate these effects, we analyze model behavior at both the token and parameter levels. Our analysis reveals that up to 90% of parameter updates during SFT do not contribute to knowledge enhancement. Restoring these updates can improve performance on the CBQA task, depending on the characteristics of the fine-tuning data. These insights offer practical guidance for developing fine-tuning strategies that more effectively strengthen model knowledge.