Large language models (LLMs) are prone to hallucinations and sensitive to prompt perturbations, often resulting in inconsistent or unreliable generated text. Different methods have been proposed to mitigate such hallucinations and fragility, one of which is to measure the consistency of LLM responses–the model’s confidence in the response or likelihood of generating a similar response when resampled. In previous work, measuring LLM response consistency often relied on calculating the probability of a response appearing within a pool of resampled responses, analyzing internal states, or evaluating logits of resopnses. However, it was not clear how well these approaches approximated users’ perceptions of consistency of LLM responses. To find out, we performed a user study (n=2,976) demonstrating that current methods for measuring LLM response consistency typically do not align well with humans’ perceptions of LLM consistency. We propose a logit-based ensemble method for estimating LLM consistency and show that our method matches the performance of the best-performing existing metric in estimating human ratings of LLM consistency. Our results suggest that methods for estimating LLM consistency without human evaluation are sufficiently imperfect to warrant broader use of evaluation with human input; this would avoid misjudging the adequacy of models because of the imperfections of automated consistency metrics.
论文检索
输入标题、作者或关键词，从 7,591 篇学术成果中精准定位
Distinguishing between rhetorical questions and informational questions is a challenging task, as many rhetorical questions have similar surface forms to informational questions. Existing datasets, however, do not contain many questions that can be rhetorical or informational in different contexts. We introduce Studying Rhetorically Ambiguous Questions (SRAQ), a new dataset explicitly constructed to support the study of such rhetorical ambiguity. The questions in SRAQ can be interpreted as either rhetorical or informational depending on the context. We evaluate the performance of state-of-the-art language models on this dataset and find that they struggle to recognize many rhetorical questions.
This paper investigates to what extent the integration of morphological information can improve subword tokenization and thus also language modeling performance. We focus on Spanish, a language with fusional morphology, where subword segmentation can benefit from linguistic structure. Instead of relying on purely data-driven strategies like Byte Pair Encoding (BPE), we explore a linguistically grounded approach: training a tokenizer on morphologically segmented data. To do so, we develop a semi-supervised segmentation model for Spanish, building gold-standard datasets to guide and evaluate it. We then use this tokenizer to pre-train a masked language model and assess its performance on several downstream tasks. Our results show improvements over a baseline with a standard tokenizer, supporting our hypothesis that morphology-aware tokenization offers a viable and principled alternative for improving language modeling.
Simple Yet Effective: An Information-Theoretic Approach to Multi-LLM Uncertainty Quantification
PDF ↗Large language models (LLMs) often behave inconsistently across inputs, indicating uncertainty and motivating the need for its quantification in high-stakes settings. Prior work on calibration and uncertainty quantification often focuses on individual models, overlooking the potential of model diversity. We hypothesize that LLMs make complementary predictions due to differences in training and the Zipfian nature of language, and that aggregating their outputs leads to more reliable uncertainty estimates. To leverage this, we propose MUSE (Multi-LLM Uncertainty via Subset Ensembles), a simple information-theoretic method that uses Jensen-Shannon Divergence to identify and aggregate well-calibrated subsets of LLMs. Experiments on binary prediction tasks demonstrate improved calibration and predictive performance compared to single-model and naive ensemble baselines. In addition, we explore using MUSE as guided signals with chain-of-thought distillation to fine-tune LLMs for calibration. MUSE is available at: https://github.com/LARK-NLP-Lab/MUSE.
Reflective Agreement: Combining Self-Mixture of Agents with a Sequence Tagger for Robust Event Extraction
PDF ↗Event Extraction (EE) involves automatically identifying and extracting structured information about events from unstructured text, including triggers, event types, and arguments. Traditional discriminative models demonstrate high precision but often exhibit limited recall, particularly for nuanced or infrequent events. Conversely, generative approaches leveraging Large Language Models (LLMs) provide higher semantic flexibility and recall but suffer from hallucinations and inconsistent predictions. To address these challenges, we propose Agreement-based Reflective Inference System (ARIS), a hybrid approach combining a Self Mixture of Agents with a discriminative sequence tagger. ARIS explicitly leverages structured model consensus, confidence-based filtering, and an LLM reflective inference module to reliably resolve ambiguities and enhance overall event prediction quality. We further investigate decomposed instruction fine-tuning for enhanced LLM event extraction understanding. Experiments demonstrate our approach outperforms existing state-of-the-art event extraction methods across three benchmark datasets.
Recent advancements in large language models have significantly influenced the field of online medical consultations. However, critical challenges remain, such as the generation of hallucinated information and the integration of up-to-date medical knowledge. To address these issues, we propose **I**nformatics **Llama** (ILlama), a novel framework that combines retrieval-augmented generation with a structured medical knowledge graph. ILlama incorporates relevant medical knowledge by transforming subgraphs from a structured medical knowledge graph into text for retrieval-augmented generation. By generating subgraphs from the medical knowledge graph in advance, specifically focusing on diseases and symptoms, ILlama is able to enhance the accuracy and relevance of its medical reasoning. This framework enables effective incorporation of causal relationships between symptoms and diseases. Also, it delivers context-aware consultations aligned with user queries. Experimental results on the two medical consultation datasets demonstrate that ILlama outperforms the strong baselines, achieving a semantic similarity F1-score of 0.884 when compared with ground truth consultation answers. Furthermore, qualitative analysis of ILlama’s responses reveals significant improvements in hallucination reduction and clinical usefulness. These results suggest that ILlama has strong potential as a reliable tool for real-world medical consultation environments.
Quantization enables efficient deployment of large language models (LLMs) in resource-constrained environments by significantly reducing memory and computation costs. While quantized LLMs often maintain performance on perplexity and zero-shot tasks, their impact on truthfulness—whether generating truthful or deceptive responses—remains largely unexplored. In this work, we introduce TruthfulnessEval, a comprehensive evaluation framework for assessing the truthfulness of quantized LLMs across three dimensions: (1) Truthfulness on Logical Reasoning; (2) Truthfulness on Common Sense; and (3) Truthfulness on Imitative Falsehoods. Using this framework, we examine mainstream quantization techniques (ranging from 4-bit to extreme 2-bit) across several open-source LLMs. Surprisingly, we find that while quantized models retain internally truthful representations, they are more susceptible to producing false outputs under misleading prompts. To probe this vulnerability, we test 15 rephrased variants of “honest”, “neutral” and “deceptive” prompts and observe that “deceptive” prompts can override truth-consistent behavior, whereas “honest” and “neutral” prompts maintain stable outputs. Further, we reveal that quantized models “know” the truth internally yet still produce false outputs when guided by “deceptive” prompts via layer-wise probing and PCA visualizations. Our findings provide insights into future designs of quantization-aware alignment and truthfulness interventions.
Innovators often exhibit cognitive fixation on existing solutions or nascent ideas, hindering the exploration of novel alternatives. This paper introduces a methodology for constructing Functional Concept Graphs (FCGs), interconnected representations of functional elements that support abstraction, problem reframing, and analogical inspiration. Our approach yields large-scale, high-quality FCGs with explicit abstraction relations, overcoming limitations of prior work. We further present MUSE, an algorithm leveraging FCGs to generate creative inspirations for a given problem. We demonstrate our method by computing an FCG on 500K patents, which we release for further research. We introduced MUSE, a novel engine to find unexpected solutions to problems. This engine consists of the inspiration graph, whose problem and solution nodes were extracted from 500K patent descriptions. For a given problem, MUSE aims to enhance users’ creative problem solving by providing them with inspirations sampled from the inspiration graph. A user study indicates that participants exposed to MUSE’s inspirations generated more creative ideas, both in terms of absolute number (up to 19% increase over participants not given inspirations) and ratio (75%, compared to 49% for no inspirations).
Despite long-standing efforts in accelerating scientific discovery with AI, building AI co-scientists remains challenging due to limited high-quality data for training and evaluation. To tackle this data scarcity issue, we present AutoSDT, an automatic pipeline that collects high-quality coding tasks in real-world data-driven discovery workflows. AutoSDT leverages the coding capabilities and parametric knowledge of LLMs to search for diverse sources, select ecologically valid tasks, and synthesize accurate task instructions and code solutions. Using our pipeline, we construct AutoSDT-5K, a dataset of 5,404 coding tasks for data-driven discovery that covers four scientific disciplines and 756 unique Python packages. To the best of our knowledge, AutoSDT-5K is the only automatically collected and the largest open dataset for data-driven scientific discovery. Expert feedback on a subset of 256 tasks shows the effectiveness of AutoSDT: 93% of the collected tasks are ecologically valid, and 92.2% of the synthesized programs are functionally correct. Trained on AutoSDT-5K, the Qwen2.5-Coder-Instruct LLM series, dubbed AutoSDT-Coder, show substantial improvement on two challenging data-driven discovery benchmarks, ScienceAgentBench and DiscoveryBench. Most notably, AutoSDT-Coder-32B reaches the same level of performance as GPT-4o on ScienceAgentBench with a success rate of 7.8%, doubling the performance of its base model. On DiscoveryBench, it lifts the hypothesis matching score to 8.1, bringing a 17.4% relative improvement and closing the gap between open-weight models and GPT-4o.
Transformers often struggle to generalize to longer sequences than those seen during training - a limitation known as length extrapolation. Most existing Relative Positional Encoding (RPE) methods attempt to address this by introducing either fixed linear biases or globally learned biases, which lack the capacity to adapt to different input contexts. In this work, we propose an additive RPE, Context-Aware Biases for Length Extrapolation (CABLE), a method that learns token-specific, context-aware biases for each attention head in transformers. By dynamically adjusting positional biases based on the input sequence, CABLE overcomes the rigidity of fixed RPEs. When evaluated on sequences longer than originally trained with, GPT-2 Medium (334M parameters) with CABLE achieves lower perplexity than counterparts using other widely adopted positional encoding methods. Additionally, by applying CABLE to the BERT base model we improved performance in long-context retrieval tasks. Our method significantly enhances the extrapolation performance of existing RPE methods tested on the FineWeb-Edu-10B and WikiText-103 datasets. Our code is available at: https://github.com/AlgonetLabs/Cable.
Enhancing Study-Level Inference from Clinical Trial Papers via Reinforcement Learning-Based Numeric Reasoning
PDF ↗Systematic reviews in medicine play a critical role in evidence-based decision-making by aggregating findings from multiple studies. A central bottleneck in automating this process is extracting numeric evidence and determining study-level conclusions for specific outcomes and comparisons. Prior work has framed this problem as a textual inference task by retrieving relevant content fragments and inferring conclusions from them. However, such approaches often rely on shallow textual cues and fail to capture the underlying numeric reasoning behind expert assessments.In this work, we conceptualise the problem as one of quantitative reasoning. Rather than inferring conclusions from surface text, we extract structured numerical evidence (e.g., event counts or standard deviations) and apply domain knowledge informed logic to derive outcome-specific conclusions. We develop a numeric reasoning system composed of a numeric data extraction model and an effect estimate component, enabling more accurate and interpretable inference aligned with the domain expert principles. We train the numeric data extraction model using different strategies, including supervised fine-tuning (SFT) and reinforcement learning (RL) with a new value reward model.When evaluated on the CochraneForest benchmark, our best-performing approach – using RL to train a small-scalenumber extraction model – yields up to a 21% absolute improvement in F1 score over retrieval-based systems and outperforms general-purpose LLMs of over 400B parameters by up to 9%.Our results demonstrate the promise of reasoning-driven approaches for automating systematic evidence synthesis.
Large Language Models (LLMs) are prone to hallucination, particularly in long-form generations. A promising direction to mitigate hallucination is to teach LLMs to express uncertainty explicitly when they lack sufficient knowledge. However, existing work lacks direct and fair evaluation of LLMs’ ability to express uncertainty effectively in long-form generation. To address this gap, we first introduce UNCLE, a benchmark designed to evaluate uncertainty expression in both long- and short-form question answering (QA). UNCLE covers five domains and includes more than 1,000 entities, each with paired short- and long-form QA items. Our dataset is the first to directly link short- and long-form QA through aligned questions and gold-standard answers.Along with UNCLE, we propose a suite of new metrics to assess the models’ capabilities to selectively express uncertainty. We then demonstrate that current models fail to convey uncertainty appropriately in long-form generation. We further explore both prompt-based and training-based methods to improve models’ performance, with the training-based methods yielding greater gains. Further analysis of alignment gaps between short- and long-form uncertainty expression highlights promising directions for future research using UNCLE.
Reinforcement learning with verifiable rewards (RLVR) has become a key technique for enhancing large language models (LLMs), with verification engineering playing a central role. However, best practices for RL in instruction following remain underexplored. In this work, we explore the verification challenge in RL for instruction following and propose VerIF, a verification method that combines rule-based code verification with LLM-based verification from a large reasoning model (e.g., QwQ-32B). To support this approach, we construct a high-quality instruction-following dataset, VerInstruct, containing approximately 22,000 instances with associated verification signals. We apply RL training with VerIF to two models, achieving significant improvements across several representative instruction-following benchmarks. The trained models reach state-of-the-art performance among models of comparable size and generalize well to unseen constraints. We further observe that their general capabilities remain unaffected, suggesting that RL with VerIF can be integrated into existing RL recipes to enhance overall model performance. We will release our datasets, codes, and models to facilitate future research.
Large language models (LLMs) are increasingly deployed in domains requiring moral understanding, yet their reasoning often remains shallow, and misaligned with human reasoning. Unlike humans, whose moral reasoning integrates contextual trade-offs, value systems, and ethical theories, LLMs often rely on surface patterns, leading to biased decisions in morally and ethically complex scenarios. To address this gap, we present a value-grounded framework for evaluating and distilling structured moral reasoning in LLMs. We benchmark 12 open-source models across four moral datasets using a taxonomy of prompts grounded in value systems, ethical theories, and cognitive reasoning strategies. Our evaluation is guided by four questions: (1) Does reasoning improve LLM decision-making over direct prompting? (2) Which types of value/ethical frameworks most effectively guide LLM reasoning? (3) Which cognitive reasoning strategies lead to better moral performance? (4) Can small-sized LLMs acquire moral competence through distillation? We find that prompting with explicit moral structure consistently improves accuracy and coherence, with first-principles reasoning and Schwartz’s + care-ethics scaffolds yielding the strongest gains. Furthermore, our supervised distillation approach transfers moral competence from large to small models without additional inference cost. Together, our results offer a scalable path toward interpretable and value-grounded models.
FLUID QA: A Multilingual Benchmark for Figurative Language Usage in Dialogue across English, Chinese, and Korean
PDF ↗Figurative language conveys stance, emotion, and social nuance, making its appropriate use essential in dialogue. While large language models (LLMs) often succeed in recognizing figurative expressions at the sentence level, their ability to use them coherently in conversation remains uncertain. We introduce FLUID QA, the first multilingual benchmark that evaluates figurative usage in dialogue across English, Korean, and Chinese. Each item embeds figurative choices into multi-turn contexts. To support interpretation, we include FLUTE-bi, a sentence-level diagnostic task. Results reveal a persistent gap: models that perform well on FLUTE-bi frequently fail on FLUID QA, especially in sarcasm and metaphor. These errors reflect systematic rhetorical confusion and limited discourse reasoning. FLUID QA provides a scalable framework for assessing usage-level figurative competence across languages.
The proliferation of multimodal content on social media presents significant challenges in understanding and moderating complex, context-dependent issues such as misinformation, hate speech, and propaganda. While efforts have been made to develop resources and propose new methods for automatic detection, limited attention has been given to label detection and the generation of explanation-based rationales for predicted labels. To address this challenge, we introduce MemeXplain, an explanation-enhanced dataset for propaganda memes in Arabic and hateful memes in English, making it the first large-scale resource for these tasks. To solve these tasks, we propose a novel multi-stage optimization approach and train Vision-Language Models (VLMs). Our results demonstrate that this approach significantly improves performance over the base model for both label detection and explanation generation, outperforming the current state-of-the-art with an absolute improvement of approximately 3% on ArMeme and 7% on Hateful Memes. For reproducibility and future research, we aim to make the MemeXplain dataset and scripts publicly available.
Modern GPUs evolve rapidly, yet production compilers still rely on hand-crafted register allocation heuristics that require substantial re-tuning for each hardware generation. We introduce VeriLocc, a framework that combines large language models (LLMs) with formal compiler techniques to enable generalizable and verifiable register allocation across GPU architectures. VeriLocc fine-tunes an LLM to translate intermediate representations (MIRs) into target-specific register assignments, aided by static analysis for cross-architecture normalization and generalization and a verifier-guided regeneration loop to ensure correctness. Evaluated on matrix multiplication (GEMM) and multi-head attention (MHA), VeriLocc achieves 85–99% single-shot accuracy and near-100% pass@100. Case study shows that VeriLocc discovers more performant assignments than expert-tuned libraries, outperforming rocBLAS by over 10% in runtime.
Disinformation poses a significant threat to democratic societies, public health, and national security. To address this challenge, fact-checking experts analyze and track disinformation narratives. However, the process of manually identifying these narratives is highly time-consuming and resource-intensive. In this article, we introduce DiNaM, the first algorithm and structured framework specifically designed for mining disinformation narratives. DiNaM uses a multi-step approach to uncover disinformation narratives. It first leverages Large Language Models (LLMs) to detect false information, then applies clustering techniques to identify underlying disinformation narratives. We evaluated DiNaM’s performance using ground-truth disinformation narratives from the EUDisinfoTest dataset. The evaluation employed the Weighted Chamfer Distance (WCD), which measures the similarity between two sets of embeddings: the ground truth and the predicted disinformation narratives. DiNaM achieved a state-of-the-art WCD score of 0.73, outperforming general-purpose narrative mining methods by a notable margin of 16.4–24.7%. We are releasing DiNaM’s codebase and the dataset to the public.
Taxonomies play a crucial role in helping researchers structure and navigate knowledge in a hierarchical manner. They also form an important part in the creation of comprehensive literature surveys. The existing approaches to automatic survey generation do not compare the structure of the generated surveys with those written by human experts. To address this gap, we present our own method for automated taxonomy creation that can bridge the gap between human-generated and automatically-created taxonomies. For this purpose, we create the CS-TaxoBench benchmark which consists of 460 taxonomies that have been extracted from human-written survey papers. We also include an additional test set of 80 taxonomies curated from conference survey papers. We propose TaxoAlign, a three-phase topic-based instruction-guided method for scholarly taxonomy generation. Additionally, we propose a stringent automated evaluation framework that measures the structural alignment and semantic coherence of automatically generated taxonomies in comparison to those created by human experts. We evaluate our method and various baselines on CS-TaxoBench, using both automated evaluation metrics and human evaluation studies. The results show that TaxoAlign consistently surpasses the baselines on nearly all metrics. The code and data can be found at https://github.com/AvishekLahiri/TaxoAlign.
Despite Greece’s pivotal role in the global economy, large language models (LLMs) remain underexplored for Greek financial context due to the linguistic complexity of Greek and the scarcity of domain-specific datasets. While multilingual financial NLP has revealed large performance gaps across languages, no benchmarks or LLMs have been tailored for Greek financial tasks until now. To bridge this gap, we introduce Plutus-ben, the first Greek Financial Evaluation Benchmark, and Plutus-8B, the first financial LLM fine-tuned on Greek-specific financial data. Plutus-ben addresses six core tasks: numeric/textual named entity recognition, question answering, extractive summarization, abstractive summarization, and topic classification. To support these tasks, we release four new expert-annotated Greek financial datasets and incorporate two existing resources. Our comprehensive evaluation of 24 LLMs reveals persistent challenges in Greek financial NLP, driven by linguistic complexity, domain terminology, and financial reasoning gaps. Experiment results underscore the limitations of cross-lingual transfer and the need for Greek-specific financial modeling. We publicly release Plutus-ben, Plutus-8B, and all associated datasets to promote reproducible research and advance multilingual financial NLP.