Reasoning with Confidence: Efficient Verification of LLM Reasoning Steps via Uncertainty Heads

Solving complex tasks usually requires LLMs to generate long multi-step reasoning chains. Previous work has shown that verifying the correctness of individual reasoning steps can further improve the performance and efficiency of LLMs on such tasks and enhance solution interpretability. However, existing verification approaches, such as Process Reward Models (PRMs), are either computationally expensive, limited to specific domains, or require large-scale human or model-generated annotations. Thus, we propose a lightweight alternative for step-level reasoning verification based on data-driven uncertainty scores. We train transformer-based uncertainty quantification heads (UHeads) that use the internal states of a frozen LLM to estimate the uncertainty of its reasoning steps during generation. The approach is fully automatic: target labels are generated either by another larger LLM (e.g., DeepSeek R1) or in a self-supervised manner by the original model itself. UHeads are both effective and lightweight, containing less than 10M parameters. Across multiple domains, including mathematics, planning, and general knowledge question answering, they match or even surpass the performance of PRMs that are up to 810x larger. Our findings suggest that the internal states of LLMs encode their uncertainty and can serve as reliable signals for reasoning verification, offering a promising direction toward scalable and generalizable introspective LLMs.

Who Evaluates AI's Social Impacts? Mapping Coverage and Gaps in First and Third Party Evaluations

Foundation models are increasingly central to high-stakes AI systems, and governance frameworks now depend on evaluations to assess their risks and capabilities. Although general capability evaluations are widespread, social impact assessments covering bias, fairness, privacy, environmental costs, and labor practices remain uneven across the AI ecosystem. To characterize this landscape, we conduct the first comprehensive analysis of both first-party and third-party social impact evaluation reporting across a wide range of model developers. Our study examines 186 first-party release reports and 183 post-release evaluation sources, and complements this quantitative analysis with interviews of model developers. We find a clear division of evaluation labor: first-party reporting is sparse, often superficial, and has declined over time in key areas such as environmental impact and bias, while third-party evaluators including academic researchers, nonprofits, and independent organizations provide broader and more rigorous coverage of bias, harmful content, and performance disparities. However, this complementarity has limits. Only model developers can authoritatively report on data provenance, content moderation labor, financial costs, and training infrastructure, yet interviews reveal that these disclosures are often deprioritized unless tied to product adoption or regulatory compliance. Our findings indicate that current evaluation practices leave major gaps in assessing AI's societal impacts, highlighting the urgent need for policies that promote developer transparency, strengthen independent evaluation ecosystems, and create shared infrastructure to aggregate and compare third-party evaluations in a consistent and accessible way.

Chimera: Diagnosing Shortcut Learning in Visual-Language Understanding

Diagrams convey symbolic information in a visual format rather than a linear stream of words, making them especially challenging for AI models to process. While recent evaluations suggest that vision-language models (VLMs) perform well on diagram-related benchmarks, their reliance on knowledge, reasoning, or modality shortcuts raises concerns about whether they genuinely understand and reason over diagrams. To address this gap, we introduce Chimera, a comprehensive test suite comprising 7,500 high-quality diagrams sourced from Wikipedia; each diagram is annotated with its symbolic content represented by semantic triples along with multi-level questions designed to assess four fundamental aspects of diagram comprehension: entity recognition, relation understanding, knowledge grounding, and visual reasoning. We use Chimera to measure the presence of three types of shortcuts in visual question answering: (1) the visual-memorization shortcut, where VLMs rely on memorized visual patterns; (2) the knowledge-recall shortcut, where models leverage memorized factual knowledge instead of interpreting the diagram; and (3) the Clever-Hans shortcut, where models exploit superficial language patterns or priors without true comprehension. We evaluate 15 open-source VLMs from 7 model families on Chimera and find that their seemingly strong performance largely stems from shortcut behaviors: visual-memorization shortcuts have slight impact, knowledge-recall shortcuts play a moderate role, and Clever-Hans shortcuts contribute significantly. These findings expose critical limitations in current VLMs and underscore the need for more robust evaluation protocols that benchmark genuine comprehension of complex visual inputs (e.g., diagrams) rather than question-answering shortcuts.

LEXam: Benchmarking Legal Reasoning on 340 Law Exams

Long-form legal reasoning remains a key challenge for large language models (LLMs) in spite of recent advances in test-time scaling. We introduce LEXam, a novel benchmark derived from 340 law exams spanning 116 law school courses across a range of subjects and degree levels. The dataset comprises 4,886 law exam questions in English and German, including 2,841 long-form, open-ended questions and 2,045 multiple-choice questions. Besides reference answers, the open questions are also accompanied by explicit guidance outlining the expected legal reasoning approach such as issue spotting, rule recall, or rule application. Our evaluation on both open-ended and multiple-choice questions present significant challenges for current LLMs; in particular, they notably struggle with open questions that require structured, multi-step legal reasoning. Moreover, our results underscore the effectiveness of the dataset in differentiating between models with varying capabilities. Adopting an LLM-as-a-Judge paradigm with rigorous human expert validation, we demonstrate how model-generated reasoning steps can be evaluated consistently and accurately. Our evaluation setup provides a scalable method to assess legal reasoning quality beyond simple accuracy metrics. Project page: https://lexam-benchmark.github.io/

Ev2R: Evaluating Evidence Retrieval in Automated Fact-Checking

Current automated fact-checking (AFC) approaches commonly evaluate evidence either implicitly via the predicted verdicts or by comparing retrieved evidence with a predefined closed knowledge source, such as Wikipedia. However, these methods suffer from limitations, resulting from their reliance on evaluation metrics developed for different purposes and constraints imposed by closed knowledge sources. Recent advances in natural language generation (NLG) evaluation offer new possibilities for evidence assessment. In this work, we introduce Ev2R, an evaluation framework for AFC that comprises three types of approaches for evidence evaluation: reference-based, proxy-reference, and reference-less. We evaluate their effectiveness through agreement with human ratings and adversarial tests, and demonstrate that prompt-based scorers, particularly those leveraging LLMs and reference evidence, outperform traditional evaluation approaches.

The Automated Verification of Textual Claims (AVeriTeC) Shared Task

The Automated Verification of Textual Claims (AVeriTeC) shared task asks participants to retrieve evidence and predict veracity for real-world claims checked by fact-checkers. Evidence can be found either via a search engine, or via a knowledge store provided by the organisers. Submissions are evaluated using AVeriTeC score, which considers a claim to be accurately verified if and only if both the verdict is correct and retrieved evidence is considered to meet a certain quality threshold. The shared task received 21 submissions, 18 of which surpassed our baseline. The winning team was TUDA_MAI with an AVeriTeC score of 63%. In this paper we describe the shared task, present the full results, and highlight key takeaways from the shared task.

TANQ: An open domain dataset of table answered questions

Language models, potentially augmented with tool usage such as retrieval are becoming the go-to means of answering questions. Understanding and answering questions in real-world settings often requires retrieving information from different sources, processing and aggregating data to extract insights, and presenting complex findings in form of structured artifacts such as novel tables, charts, or infographics. In this paper, we introduce TANQ, the first open domain question answering dataset where the answers require building tables from information across multiple sources. We release the full source attribution for every cell in the resulting table and benchmark state-of-the-art language models in open, oracle, and closed book setups. Our best-performing baseline, GPT4 reaches an overall F1 score of 29.1, lagging behind human performance by 19.7 points. We analyse baselines' performance across different dataset attributes such as different skills required for this task, including multi-hop reasoning, math operations, and unit conversions. We further discuss common failures in model-generated answers, suggesting that TANQ is a complex task with many challenges ahead.

Croissant: A Metadata Format for ML-Ready Datasets

Data is a critical resource for Machine Learning (ML), yet working with data remains a key friction point. This paper introduces Croissant, a metadata format for datasets that simplifies how data is used by ML tools and frameworks. Croissant makes datasets more discoverable, portable and interoperable, thereby addressing significant challenges in ML data management and responsible AI. Croissant is already supported by several popular dataset repositories, spanning hundreds of thousands of datasets, ready to be loaded into the most popular ML frameworks.

ChartCheck: An Evidence-Based Fact-Checking Dataset over Real-World Chart Images

Data visualizations are common in the real-world. We often use them in data sources such as scientific documents, news articles, textbooks, and social media to summarize key information in a visual form. Charts can also mislead its audience by communicating false information or biasing them towards a specific agenda. Verifying claims against charts is not a straightforward process. It requires analyzing both the text and visual components of the chart, considering characteristics such as colors, positions, and orientations. Moreover, to determine if a claim is supported by the chart content often requires different types of reasoning. To address this challenge, we introduce ChartCheck, a novel dataset for fact-checking against chart images. ChartCheck is the first large-scale dataset with 1.7k real-world charts and 10.5k human-written claims and explanations. We evaluated the dataset on state-of-the-art models and achieved an accuracy of 73.9 in the finetuned setting. Additionally, we identified chart characteristics and reasoning types that challenge the models.

Exploring the Numerical Reasoning Capabilities of Language Models: A Comprehensive Analysis on Tabular Data

Numbers are crucial for various real-world domains such as finance, economics, and science. Thus, understanding and reasoning with numbers are essential skills for language models to solve different tasks. While different numerical benchmarks have been introduced in recent years, they are limited to specific numerical aspects mostly. In this paper, we propose a hierarchical taxonomy for numerical reasoning skills with more than ten reasoning types across four levels: representation, number sense, manipulation, and complex reasoning. We conduct a comprehensive evaluation of state-of-the-art models to identify reasoning challenges specific to them. Henceforth, we develop a diverse set of numerical probes employing a semi-automated approach. We focus on the tabular Natural Language Inference (TNLI) task as a case study and measure models' performance shifts. Our results show that no model consistently excels across all numerical reasoning types. Among the probed models, FlanT5 (few-/zero-shot) and GPT-3.5 (few-shot) demonstrate strong overall numerical reasoning skills compared to other models. Label-flipping probes indicate that models often exploit dataset artifacts to predict the correct labels.