Existing research assesses LLMs’ values by analyzing their stated inclinations, overlooking potential discrepancies between stated values and actions—termed the “Value-Action Gap.” This study introduces ValueActionLens, a framework to evaluate the alignment between LLMs’ stated values and their value-informed actions. The framework includes a dataset of 14.8k value-informed actions across 12 cultures and 11 social topics, along with two tasks measuring alignment through three metrics. Experiments show substantial misalignment between LLM-generated value statements and their actions, with significant variations across scenarios and models. Misalignments reveal potential harms, highlighting risks in relying solely on stated values to predict behavior. The findings stress the need for context-aware evaluations of LLM values and the value-action gaps.
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
LLMs as World Models: Data-Driven and Human-Centered Pre-Event Simulation for Disaster Impact Assessment
PDF ↗Efficient simulation is essential for enhancing proactive preparedness for sudden-onset disasters such as earthquakes. Recent advancements in large language models (LLMs) as world models show promise in simulating complex scenarios. This study examines multiple LLMs to proactively estimate perceived earthquake impacts. Leveraging multimodal datasets including geospatial, socioeconomic, building, and street-level imagery data, our framework generates Modified Mercalli Intensity (MMI) predictions at zip code and county scales. Evaluations on the 2014 Napa and 2019 Ridgecrest earthquakes using USGS “Did You Feel It? (DYFI)” reports demonstrate significant alignment, as evidenced by high correlation of 0.88 and low RMSE of 0.77 as compared to real reports at the zip code level. Techniques such as RAG and ICL can improve simulation performance, while visual inputs notably enhance accuracy compared to structured numerical data alone. These findings show the promise of LLMs in simulating disaster impacts that can help strengthen pre-event planning.
Large Language Model (LLM) agents are commonly tuned with supervised finetuning on ReAct-style expert trajectories or preference optimization over pairwise rollouts. Most of these methods focus on imitating specific expert behaviors or promoting chosen reasoning thoughts and actions over rejected ones. However, without reasoning and comparing over alternatives actions, LLM agents finetuned with these methods may over-commit towards seemingly plausible but suboptimal actions due to limited action space exploration. To address this, in this paper we propose Self-taught ActioN Deliberation (SAND) framework, enabling LLM agents to explicitly deliberate over candidate actions before committing to one. To tackle the challenges of when and what to deliberate given large action space and step-level action evaluation, we incorporate self-consistency action sampling and execution-guided action critique to help synthesize step-wise action deliberation thoughts using the base model of the LLM agent. In an iterative manner, the deliberation trajectories are then used to finetune the LLM agent itself. Evaluating on two representative interactive agent tasks, SAND achieves an average 20% improvement over initial supervised finetuning and also outperforms state-of-the-art agent tuning approaches.
Autoregressive generative models play a key role in various language tasks, especially for modeling and evaluating long text sequences. While recent methods leverage stochastic representations to better capture sequence dynamics, encoding both temporal and structural dependencies and utilizing such information for evaluation remains challenging. In this work, we observe that fitting transformer-based model embeddings into a stochastic process yields ordered latent representations from originally unordered model outputs. Building on this insight and prior work, we theoretically introduce a novel likelihood-based evaluation metric BBScoreV2. Empirically, we demonstrate that the stochastic latent space induces a “clustered-to-temporal ordered” mapping of language model representations in high-dimensional space, offering both intuitive and quantitative support for the effectiveness of BBScoreV2. Furthermore, this structure aligns with intrinsic properties of natural language and enhances performance on tasks such as temporal consistency evaluation (e.g., Shuffle tasks) and AI-generated content detection.
Recent decoding methods improve the factuality of large language models (LLMs) by refining how the next token is selected during generation. These methods typically operate at the token level, leveraging internal representations to suppress superficial patterns. Nevertheless, LLMs remain prone to hallucinations, especially over longer contexts. In this paper, we propose Active Layer-Contrastive Decoding (ActLCD), a novel decoding strategy that actively decides when to apply contrasting layers during generation. By casting decoding as a sequential decision-making problem, ActLCD employs a reinforcement learning policy guided by a reward-aware classifier to optimize factuality beyond the token level. Our experiments demonstrate that ActLCD surpasses state-of-the-art methods across five benchmarks, showcasing its effectiveness in mitigating hallucinations in diverse generation scenarios.
Critique Fine-Tuning (CFT) has recently emerged as a promising paradigm for unlocking the reasoning capabilities of large language models (LLMs). In this work, we introduce one-shot CFT, a highly compute-efficient approach that leverages critique data generated from a single math problem. Remarkably, this method yields significant gains in reasoning accuracy, surpassing one-shot RLVR (Reinforcement Learning with Verifiable Reward) while requiring 15 to 20 times less compute. Given one math problem, we first prompt a set of diverse small models to produce candidate solutions, then use frontier models such as GPT-4.1 to generate high-quality critiques of these responses. We fine-tune Qwen and Llama family models ranging from 1.5B to 14B parameters with CFT. With just 5 GPU hours, our models achieve up to a 16 percent absolute improvement in average accuracy across six mathematical reasoning benchmarks (for example, Qwen2.5-Math-7B improves from 26 percent to 42 percent). Furthermore, ablation studies reveal the robustness of one-shot CFT across different prompt problems. Our findings suggest an extremely compute-efficient approach to unleash the reasoning potential of LLMs.
Encoder-decoder models have achieved remarkable success in speech and text tasks, yet efficiently adapting these models to diverse uni/multi-modal scenarios remains an open challenge. In this paper, we propose Whisper-UT, a unified and efficient framework that leverages lightweight adapters to enable seamless adaptation across tasks, including a multi-modal machine translation (MMT) task that explicitly conditions translation on both speech and source language text inputs. By incorporating ASR hypotheses or ground-truth transcripts as prompts, this approach not only enables the system to process both modalities simultaneously but also enhances speech translation (ST) performance through a 2-stage decoding strategy. We demonstrate our methods using the Whisper model, though in principle they are general and could be applied to similar multitask models. We highlight the effectiveness of cross-modal and cross-task fine-tuning, which improves performance without requiring 3-way parallel data. Our results underscore the flexibility, efficiency, and general applicability of the proposed framework for multi-modal translation.
Modern language models (LMs) are not robust to out-of-distribution inputs. Machine generated (“optimized”) prompts can be used to modulate LM outputs and induce specific behaviors while appearing completely uninterpretable. In this work, we investigate the composition of optimized prompts, as well as the mechanisms by which LMs parse and build predictions from optimized prompts. We find that optimized prompts primarily consist of punctuation and noun tokens which are more rare in the training data. Internally, optimized prompts are clearly distinguishable from natural language counterparts based on sparse subsets of the model’s activations. Across various families of instruction-tuned models, optimized prompts follow a similar path in how their representations form through the network.
Understanding the prevalence of misinformation in health topics online can inform public health policies and interventions. However, measuring such misinformation at scale remains a challenge, particularly for high-stakes but understudied topics like opioid-use disorder (OUD)—a leading cause of death in the U.S. We present the first large-scale study of OUD-related myths on YouTube, a widely-used platform for health information. With clinical experts, we validate 8 pervasive myths and release an expert-labeled video dataset. To scale labeling, we introduce MythTriage, an efficient triage pipeline that uses a lightweight model for routine cases and defers harder ones to a high-performing, but costlier, large language model (LLM). MythTriage achieves up to 0.86 macro F1-score while estimated to reduce annotation time and financial cost by over 76% compared to experts and full LLM labeling. We analyze 2.9K search results and 343K recommendations, uncovering how myths persist on YouTube and offering actionable insights for public health and platform moderation.
Alignment Quality Index (AQI) : Beyond Refusals: AQI as an Intrinsic Alignment Diagnostic via Latent Geometry, Cluster Divergence, and Layer wise Pooled Representations
PDF ↗Alignment is no longer a luxury; it is a necessity. As large language models (LLMs) enter high-stakes domains like education, healthcare, governance, and law, their behavior must reliably reflect human-aligned values and safety constraints. Yet current evaluations rely heavily on behavioral proxies such as refusal rates, G-Eval scores, and toxicity classifiers, all of which have critical blind spots. Aligned models are often vulnerable to jailbreaking, stochasticity of generation, and alignment faking. To address this issue, we introduce the **Alignment Quality Index (AQI)**. This novel geometric and prompt-invariant metric empirically assesses LLM alignment by analyzing the separation of safe and unsafe activations in latent space. By combining measures such as the *Davies-Bouldin score (DBS)*, *Dunn index (DI)*, *Xie-Beni index (XBI)*, and *Calinski-Harabasz index (CHI)* across various formulations, AQI captures clustering quality to detect hidden misalignments and jailbreak risks, even when outputs appear compliant. AQI also serves as an early warning signal for alignment faking, offering a robust, decoding-invariant tool for behavior-agnostic safety auditing. Additionally, we propose the **LITMUS** dataset to facilitate robust evaluation under these challenging conditions. Empirical tests on LITMUS across different models trained under DPO, GRPO, and RLHF conditions demonstrate AQI’s correlation with external judges and ability to reveal vulnerabilities missed by refusal metrics. We make our implementation publicly available to foster future research in this area.
NLP models often rely on human-labeled data for training and evaluation. Many approaches crowdsource this data from a large number of annotators with varying skills, backgrounds, and motivations, resulting in conflicting annotations. These conflicts have traditionally been resolved by aggregation methods that assume disagreements are errors. Recent work has argued that for many tasks annotators may have genuine disagreements and that variation should be treated as signal rather than noise. However, few models separate signal and noise in annotator disagreement. In this work, we introduce NUTMEG, a new Bayesian model that incorporates information about annotator backgrounds to remove noisy annotations from human-labeled training data while preserving systematic disagreements. Using synthetic and real-world data, we show that NUTMEG is more effective at recovering ground-truth from annotations with systematic disagreement than traditional aggregation methods, and we demonstrate that downstream models trained on NUTMEG-aggregated data significantly outperform models trained on data from traditionally aggregation methods. We provide further analysis characterizing how differences in subpopulation sizes, rates of disagreement, and rates of spam affect the performance of our model. Our results highlight the importance of accounting for both annotator competence and systematic disagreements when training on human-labeled data.
MedHallu: A Comprehensive Benchmark for Detecting Medical Hallucinations in Large Language Models
PDF ↗Advancements in Large Language Models (LLMs) and their increasing use in medical question-answering necessitate rigorous evaluation of their reliability. A critical challenge lies in hallucination, where models generate plausible yet factually incorrect outputs. In the medical domain, this poses serious risks to patient safety and clinical decision-making. To address this, we introduce MedHallu, one of the the first benchmark specifically designed for medical hallucination detection. MedHallu comprises 10,000 high-quality question-answer pairs derived from PubMedQA, with hallucinated answers systematically generated through a controlled pipeline. Our experiments show that state-of-the-art LLMs, including GPT-4o, Llama-3.1, and the medically fine-tuned UltraMedical, struggle with this binary hallucination detection task, with the best model achieving an F1 score as low as 0.625 for detecting ”hard” category hallucinations. Using bidirectional entailment clustering, we show that harder-to-detect hallucinations are semantically closer to ground truth. Through experiments, we also show incorporating domain-specific knowledge and introducing a ”not sure” category as one of the answer categories improves the precision and F1 scores by up to 38% relative to baselines.
Ask Patients with Patience: Enabling LLMs for Human-Centric Medical Dialogue with Grounded Reasoning
PDF ↗The severe shortage of medical doctors limits access to timely and reliable healthcare, leaving millions underserved. Large language models (LLMs) offer a potential solution but struggle in real-world clinical interactions. Many LLMs are not grounded in authoritative medical guidelines and fail to transparently manage diagnostic uncertainty. Their language is often rigid and mechanical, lacking the human-like qualities essential for patient trust. To address these challenges, we propose ***Ask Patients with Patience (APP)***, a multi-turn LLM-based medical assistant designed for grounded reasoning, transparent diagnoses, and human-centric interaction. APP enhances communication by eliciting user symptoms through empathetic dialogue, significantly improving accessibility and user engagement. It also incorporates Bayesian active learning to support transparent and adaptive diagnoses. The framework is built on verified medical guidelines, ensuring clinically grounded and evidence-based reasoning. To evaluate its performance, we develop a new benchmark that simulates realistic medical conversations using patient agents driven by profiles extracted from real-world consultation cases. We compare APP against SOTA one-shot and multi-turn LLM baselines. The results show that APP improves diagnostic accuracy, reduces uncertainty, and enhances user experience. By integrating medical expertise with transparent, human-like interaction, APP bridges the gap between AI-driven medical assistance and real-world clinical practice.
Advances in large language models (LLMs) significantly enhance reasoning capabilities but their deployment is restricted in resource-constrained scenarios. Knowledge distillation addresses this by transferring knowledge from powerful teacher models to compact and transparent students.However, effectively capturing the teacher’s comprehensive reasoning is challenging due to conventional token-level supervision’s limited scope. Using multiple reasoning paths per query alleviates this problem, but treating each path identically is suboptimal as paths vary widely in quality and suitability across tasks and models.We propose Quality-filtered Routing with Cooperative Distillation(QR-Distill), combining path quality filtering, conditional routing, and cooperative peer teaching. First, quality filtering retains only correct reasoning paths scored by an LLM-based evaluation. Second, conditional routing dynamically assigns paths tailored to each student’s current learning state. Finally, cooperative peer teaching enables students to mutually distill diverse insights, addressing knowledge gaps and biases toward specific reasoning styles. Experiments demonstrate QR-Distill’s superiority over traditional single- and multi-path distillation methods. Ablation studies further highlight the importance of each component—quality filtering, conditional routing, and peer teaching—in effective knowledge transfer. Our code is available at https://github.com/LzyFischer/Distill.
TTT-Bench: A Benchmark for Evaluating Reasoning Ability with Simple and Novel Tic-Tac-Toe-style Games
PDF ↗Large reasoning models (LRMs) have demonstrated impressive reasoning capabilities across a broad range of tasks including Olympiad-level mathematical problems, indicating evidence of their complex reasoning abilities. While many reasoning benchmarks focus on the STEM domain, the ability of LRMs to reason correctly in broader task domains remains underexplored. In this work, we introduce **TTT-Bench**, a new benchmark that is designed to evaluate basic strategic, spatial, and logical reasoning abilities in LRMs through a suite of four two-player Tic-Tac-Toe-style games that humans can effortlessly solve from a young age. We propose a simple yet scalable programmatic approach for generating verifiable two-player game problems for TTT-Bench. Although these games are trivial for humans, they require reasoning about the intentions of the opponent, as well as the game board’s spatial configurations, to ensure a win. We evaluate a diverse set of state-of-the-art LRMs, and **discover that the models that excel at hard math problems frequently fail at these simple reasoning games**. Further testing reveals that our evaluated reasoning models score on average \downarrow 41% & \downarrow 5% lower on TTT-Bench compared to MATH 500 & AIME 2024 respectively, with larger models achieving higher performance using shorter reasoning traces, where most of the models struggle on long-term strategic reasoning situations on simple and new TTT-Bench tasks.
MultiMatch: Multihead Consistency Regularization Matching for Semi-Supervised Text Classification
PDF ↗We introduce **MultiMatch**, a novel semi-supervised learning (SSL) algorithm combining the paradigms of co-training and consistency regularization with pseudo-labeling. At its core, MultiMatch features a three-fold pseudo-label weighting module designed for selecting and filtering pseudo-labels based on head agreement and model confidence, and weighting them according to the perceived classification difficulty. This novel module enhances and unifies three existing techniques - heads agreement from **Multi**head Co-training, self-adaptive thresholds from Free**Match**, and Average Pseudo-Margins from Margin**Match** - resulting in a holistic approach that improves robustness and performance in SSL settings.Experimental results on benchmark datasets highlight the superior performance of MultiMatch, i.e., MultiMatch achieves state-of-the-art results on 8 out of 10 setups from 5 natural language processing datasets and ranks first according to the Friedman test among 21 methods. Furthermore, MultiMatch demonstrates exceptional robustness in highly imbalanced settings, outperforming the second-best approach by 3.26%, a critical advantage for real-world text classification tasks. Our code is available on GitHub.
Assessment and evaluation have long been critical challenges in artificial intelligence (AI) and natural language processing (NLP). Traditional methods, usually matching-based or small model-based, often fall short in open-ended and dynamic scenarios. Recent advancements in Large Language Models (LLMs) inspire the “LLM-as-a-judge” paradigm, where LLMs are leveraged to perform scoring, ranking, or selection for various machine learning evaluation scenarios. This paper presents a comprehensive survey of LLM-based judgment and assessment, offering an in-depth overview to review this evolving field. We first provide the definition from both input and output perspectives. Then we introduce a systematic taxonomy to explore LLM-as-a-judge along three dimensions: what to judge, how to judge, and how to benchmark. Finally, we also highlight key challenges and promising future directions for this emerging area.
We introduce VoiceCraft-X, an autoregressive neural codec language model which unifies multilingual speech editing and zero-shot text-to-speech (TTS) synthesis across 11 languages: English, Mandarin, Korean, Japanese, Spanish, French, German, Dutch, Italian, Portuguese, and Polish. VoiceCraft-X utilizes the Qwen3 large language model for phoneme-free cross-lingual text processing and a novel token reordering mechanism with time-aligned text and speech tokens to handle both tasks as a single sequence generation problem. The model generates high-quality, natural-sounding speech, seamlessly creating new audio or editing existing recordings within one framework. VoiceCraft-X shows robust performance in diverse linguistic settings, even with limited per-language data, underscoring the power of unified autoregressive approaches for advancing complex, real-world multilingual speech applications. Audio samples are available at https://zhishengzheng.com/voicecraft-x/.
Automatic survey generation has emerged as a key task in scientific document processing. While large language models (LLMs) have shown promise in generating survey texts, the lack of standardized evaluation datasets critically hampers rigorous assessment of their performance against human-written surveys. In this work, we present SurveyGen, a large-scale dataset comprising over 4,200 human-written surveys across diverse scientific domains, along with 242,143 cited references and extensive quality-related metadata for both the surveys and the cited papers. Leveraging this resource, we build QUAL-SG, a novel quality-aware framework for survey generation that enhances the standard Retrieval-Augmented Generation (RAG) pipeline by incorporating quality-aware indicators into literature retrieval to assess and select higher-quality source papers. Using this dataset and framework, we systematically evaluate state-of-the-art LLMs under varying levels of human involvement—from fully automatic generation to human-guided writing. Experimental results and human evaluations show that while semi-automatic pipelines can achieve partially competitive outcomes, fully automatic survey generation still suffers from low citation quality and limited critical analysis.
COCO-Tree: Compositional Hierarchical Concept Trees for Enhanced Reasoning in Vision-Language Models
PDF ↗Compositional reasoning remains a persistent weakness of modern vision language models (VLMs): they often falter when a task hinges on understanding how multiple objects, attributes, and relations interact within an image. Multiple research works have attempted to improve compositionality performance by creative tricks such as improving prompt structure, chain of thought reasoning, etc. A more recent line of work attempts to impart additional reasoning in VLMs using well-trained Large Language Models (LLMs), which are far superior in linguistic understanding than VLMs to compensate for the limited linguistic prowess of VLMs. However, these approaches are either resource-intensive or do not provide an interpretable reasoning process. In this paper, we present “COCO-Tree” - a novel approach that augments VLM outputs with carefully designed neurosymbolic concept trees learned from LLMs to improve VLM’s linguistic reasoning. COCO-Tree’s beam search-inspired reasoning process boosts compositionality performance and provides a rationale behind VLM predictions. Empirical results on four compositionality benchmarks, Winoground, EqBench, ColorSwap, and SugarCrepe, in seven different open-source VLMs with varying sizes, demonstrate that COCO-Tree significantly improves compositional generalization by 5-10% over baselines.