There are currently two main paradigms for evaluating large language models (LLMs), reference-based evaluation and preference-based evaluation. The first, carried over from the evaluation of machine learning models in general, relies on pre-defined task instances, for which reference task executions are available. The second, best exemplified by the LM-arena, relies on (often self-selected) users bringing their own intents to a site that routes these to several models in parallel, among whose responses the user then selects their most preferred one. The former paradigm hence excels at control over what is tested, while the latter comes with higher ecological validity, testing actual use cases interactively. Recently, a third complementary paradigm has emerged that combines some of the strengths of these approaches, offering control over multi-turn, reference-free, repeatable interactions, while stressing goal-directedness: dialogue game based evaluation. While the utility of this approach has been shown by several projects, its adoption has been held back by the lack of a mature, easily re-usable implementation. In this paper, we present clembench, which has been in continuous development since 2023 and has in its latest release been optimized for ease of general use. We describe how it can be used to benchmark one's own models (using a provided set of benchmark game instances in English), as well as how easily the benchmark itself can be extended with new, tailor-made targeted tests.

Sparse autoencoders (SAEs) and transcoders have become important tools for machine learning interpretability. However, measuring how interpretable they are remains challenging, with weak consensus about which benchmarks to use. Most evaluation procedures start by producing a single-sentence explanation for each latent. These explanations are then evaluated based on how well they enable an LLM to predict the activation of a latent in new contexts. This method makes it difficult to disentangle the explanation generation and evaluation process from the actual interpretability of the latents discovered. In this work, we adapt existing methods to assess the interpretability of sparse coders, with the advantage that they do not require generating natural language explanations as an intermediate step. This enables a more direct and potentially standardized assessment of interpretability. Furthermore, we compare the scores produced by our interpretability metrics with human evaluations across similar tasks and varying setups, offering suggestions for the community on improving the evaluation of these techniques.

This paper provides an experimental evaluation of the capability of large language models (LLMs) to assist in legal decision-making within the framework of Austrian and European Union value-added tax (VAT) law. In tax consulting practice, clients often describe cases in natural language, making LLMs a prime candidate for supporting automated decision-making and reducing the workload of tax professionals. Given the requirement for legally grounded and well-justified analyses, the propensity of LLMs to hallucinate presents a considerable challenge. The experiments focus on two common methods for enhancing LLM performance: fine-tuning and retrieval-augmented generation (RAG). In this study, these methods are applied on both textbook cases and real-world cases from a tax consulting firm to systematically determine the best configurations of LLM-based systems and assess the legal-reasoning capabilities of LLMs. The findings highlight the potential of using LLMs to support tax consultants by automating routine tasks and providing initial analyses, although current prototypes are not ready for full automation due to the sensitivity of the legal domain. The findings indicate that LLMs, when properly configured, can effectively support tax professionals in VAT tasks and provide legally grounded justifications for decisions. However, limitations remain regarding the handling of implicit client knowledge and context-specific documentation, underscoring the need for future integration of structured background information.

With the widespread application of Large Language Models (LLMs) in various tasks, the mainstream LLM platforms generate massive user-model interactions daily. In order to efficiently analyze the performance of models and diagnose failures in their answers, it is essential to develop an automated framework to systematically categorize and attribute errors. However, existing evaluation models lack error attribution capability. In this work, we establish a comprehensive Misattribution Framework with 6 primary and 15 secondary categories to facilitate in-depth analysis. Based on this framework, we present AttriData, a dataset specifically designed for error attribution, encompassing misattribution, along with the corresponding scores and feedback. We also propose MisAttributionLLM, a fine-tuned model on AttriData, which is the first general-purpose judge model capable of simultaneously generating score, misattribution, and feedback. Extensive experiments and analyses are conducted to confirm the effectiveness and robustness of our proposed method.

Retrieval-Augmented Generation (RAG) enhances language models by grounding responses in external information, yet explainability remains a critical challenge, particularly when retrieval relies on unstructured text. Knowledge graphs (KGs) offer a solution by introducing structured, semantically rich representations of entities and their relationships, enabling transparent retrieval paths and interpretable reasoning. In this work, we present KGRAG-Ex, a RAG system that improves both factual grounding and explainability by leveraging a domain-specific KG constructed via prompt-based information extraction. Given a user query, KGRAG-Ex identifies relevant entities and semantic paths in the graph, which are then transformed into pseudo-paragraphs: natural language representations of graph substructures that guide corpus retrieval. To improve interpretability and support reasoning transparency, we incorporate perturbation-based explanation methods that assess the influence of specific KG-derived components on the generated answers. We conduct a series of experiments to analyze the sensitivity of the system to different perturbation methods, the relationship between graph component importance and their structural positions, the influence of semantic node types, and how graph metrics correspond to the influence of components within the explanations process.

Shapes Constraint Language (SHACL) is a powerful language for validating RDF data. Given the recent industry attention to Knowledge Graphs (KGs), more users need to validate linked data properly. However, traditional SHACL validation engines often provide terse reports in English that are difficult for non-technical users to interpret and act upon. This paper presents xpSHACL, an explainable SHACL validation system that addresses this issue by combining rule-based justification trees with retrieval-augmented generation (RAG) and large language models (LLMs) to produce detailed, multilanguage, human-readable explanations for constraint violations. A key feature of xpSHACL is its usage of a Violation KG to cache and reuse explanations, improving efficiency and consistency.

Current knowledge editing methods for large language models (LLMs) struggle to maintain logical consistency when propagating ripple effects to associated facts. We propose ChainEdit, a framework that synergizes knowledge graph-derived logical rules with LLM logical reasoning capabilities to enable systematic chain updates. By automatically extracting logical patterns from structured knowledge bases and aligning them with LLMs' internal logics, ChainEdit dynamically generates and edits logically connected knowledge clusters. Experiments demonstrate an improvement of more than 30% in logical generalization over baselines while preserving editing reliability and specificity. We further address evaluation biases in existing benchmarks through knowledge-aware protocols that disentangle external dependencies. This work establishes new state-of-the-art performance on ripple effect while ensuring internal logical consistency after knowledge editing.

Large Language Models (LLMs) have demonstrated their transformative potential across numerous disciplinary studies, reshaping the existing research methodologies and fostering interdisciplinary collaboration. However, a systematic understanding of their integration into diverse disciplines remains underexplored. This survey paper provides a comprehensive overview of the application of LLMs in interdisciplinary studies, categorising research efforts from both a technical perspective and with regard to their applicability. From a technical standpoint, key methodologies such as supervised fine-tuning, retrieval-augmented generation, agent-based approaches, and tool-use integration are examined, which enhance the adaptability and effectiveness of LLMs in discipline-specific contexts. From the perspective of their applicability, this paper explores how LLMs are contributing to various disciplines including mathematics, physics, chemistry, biology, and the humanities and social sciences, demonstrating their role in discipline-specific tasks. The prevailing challenges are critically examined and the promising research directions are highlighted alongside the recent advances in LLMs. By providing a comprehensive overview of the technical developments and applications in this field, this survey aims to serve as an invaluable resource for the researchers who are navigating the complex landscape of LLMs in the context of interdisciplinary studies.

Language models are prone to hallucination - generating text that is factually incorrect. Finetuning models on high-quality factual information can potentially reduce hallucination, but concerns remain; obtaining factual gold data can be expensive and training on correct but unfamiliar data may potentially lead to even more downstream hallucination. What data should practitioners finetune on to mitigate hallucinations in language models? In this work, we study the relationship between the factuality of finetuning data and the prevalence of hallucinations in long-form generation tasks. Counterintuitively, we find that finetuning on factual gold data is not as helpful as finetuning on model-generated data that models believe to be factual. Next, we evaluate filtering strategies applied on both factual gold data and model-generated data, and find that finetuning on model-generated data that is filtered by models' own internal judgments often leads to better overall factuality compared to other configurations: training on gold data filtered by models' judgments, training on gold data alone, or training on model-generated data that is supported by gold data. These factuality improvements transfer across three domains we study, suggesting that a models' own beliefs can provide a powerful signal for factuality.

Efficient planning, resource management, and consistent operations often rely on converting textual process documents into formal Business Process Model and Notation (BPMN) models. However, this conversion process remains time-intensive and costly. Existing approaches, whether rule-based or machine-learning-based, still struggle with writing styles and often fail to identify parallel structures in process descriptions. This paper introduces an automated pipeline for extracting BPMN models from text, leveraging the use of machine learning and large language models. A key contribution of this work is the introduction of a newly annotated dataset, which significantly enhances the training process. Specifically, we augment the PET dataset with 15 newly annotated documents containing 32 parallel gateways for model training, a critical feature often overlooked in existing datasets. This addition enables models to better capture parallel structures, a common but complex aspect of process descriptions. The proposed approach demonstrates adequate performance in terms of reconstruction accuracy, offering a promising foundation for organizations to accelerate BPMN model creation.