This paper investigates the impact of incorrect data on the performance and safety of large language models (LLMs), specifically gpt-4o, during supervised fine-tuning (SFT). Although LLMs become increasingly vital across broad domains like finance, coding, law, and health, fine-tuning on incorrect data can lead to "emergent misalignment," producing harmful or deceptive outputs unrelated to the intended task. We evaluate gpt-4o models fine-tuned with varying ratios (10\% to 90\% correct) of both obviously and subtly incorrect data across four domains: coding, finance, health, and legal. Our findings show that even modest amounts of incorrect data (10-25\%) dramatically degrade domain performance and not moral alignment. A clear threshold of at least 50\% correct data is needed for models to consistently recover strong performance, though they rarely match the robustness and safety of the base model, which exhibits near-perfect alignment and zero dangerous completions out-of-the-box. This research emphasizes that the cost of incorrect data is heavy, highlighting the critical need for extremely high-quality data curation or, alternatively, leveraging robust base models without unnecessary fine-tuning for high-stakes applications.

Despite being trained on significant amounts of data, Large Language Models (LLMs) can provide inaccurate or unreliable information in the context of a user's specific query. Given query-specific context significantly improves the usefulness of its responses. In this paper, we present a specification that can be used to dynamically build context for data sources. The data source owner creates the file containing metadata for LLMs to use when reasoning about dataset-related queries. To demonstrate our proposed specification, we created a prototype Readme_AI Model Context Protocol (MCP) server that retrieves the metadata from the data source and uses it to dynamically build context. Some features that make this specification dynamic are the extensible types that represent crawling web-pages, fetching data from data repositories, downloading and parsing publications, and general text. The context is formatted and grouped using user-specified tags that provide clear contextual information for the LLM to reason about the content. We demonstrate the capabilities of this early prototype by asking the LLM about the NIST-developed Hedgehog library, for which common LLMs often provides inaccurate and irrelevant responses containing hallucinations. With Readme_AI, the LLM receives enough context that it is now able to reason about the library and its use, and even generate code interpolated from examples that were included in the Readme_AI file provided by Hedgehog's developer. Our primary contribution is a extensible protocol for dynamically grounding LLMs in specialized, owner-provided data, enhancing responses from LLMs and reducing hallucinations. The source code for the Readme_AI tool is posted here: https://github.com/usnistgov/readme_ai .

This study evaluates item neutralization assisted by the large language model (LLM) to reduce social desirability bias in personality assessment. GPT-o3 was used to rewrite the International Personality Item Pool Big Five Measure (IPIP-BFM-50), and 203 participants completed either the original or neutralized form along with the Marlowe-Crowne Social Desirability Scale. The results showed preserved reliability and a five-factor structure, with gains in Conscientiousness and declines in Agreeableness and Openness. The correlations with social desirability decreased for several items, but inconsistently. Configural invariance held, though metric and scalar invariance failed. Findings support AI neutralization as a potential but imperfect bias-reduction method.

The use of continuous instead of discrete tokens during the Chain-of-Thought (CoT) phase of reasoning LLMs has garnered attention recently, based on the intuition that a continuous mixture of discrete tokens could simulate a superposition of several reasoning paths simultaneously. Theoretical results have formally proven that continuous tokens have much greater expressivity and can solve specific problems more efficiently. However, practical use of continuous tokens has been limited by strong training difficulties: previous works either just use continuous tokens at inference time on a pre-trained discrete-token model, or must distill the continuous CoT from ground-truth discrete CoTs and face computational costs that limit the CoT to very few tokens. This is the first work introducing a scalable method to learn continuous CoTs via reinforcement learning (RL), without distilling from reference discrete CoTs. We use "soft" tokens: mixtures of tokens together with noise on the input embedding to provide RL exploration. Computational overhead is minimal, enabling us to learn continuous CoTs with hundreds of tokens. On math reasoning benchmarks with Llama and Qwen models up to 8B, training with continuous CoTs match discrete-token CoTs for pass@1 and surpass them for pass@32, showing greater CoT diversity. In systematic comparisons, the best-performing scenario is to train with continuous CoT tokens then use discrete tokens for inference, meaning the "soft" models can be deployed in a standard way. Finally, we show continuous CoT RL training better preserves the predictions of the base model on out-of-domain tasks, thus providing a softer touch to the base model.

Medical imaging provides critical evidence for clinical diagnosis, treatment planning, and surgical decisions, yet most existing imaging models are narrowly focused and require multiple specialized networks, limiting their generalization. Although large-scale language and multimodal models exhibit strong reasoning and multi-task capabilities, real-world clinical applications demand precise visual grounding, multimodal integration, and chain-of-thought reasoning. We introduce Citrus-V, a multimodal medical foundation model that combines image analysis with textual reasoning. The model integrates detection, segmentation, and multimodal chain-of-thought reasoning, enabling pixel-level lesion localization, structured report generation, and physician-like diagnostic inference in a single framework. We propose a novel multimodal training approach and release a curated open-source data suite covering reasoning, detection, segmentation, and document understanding tasks. Evaluations demonstrate that Citrus-V outperforms existing open-source medical models and expert-level imaging systems across multiple benchmarks, delivering a unified pipeline from visual grounding to clinical reasoning and supporting precise lesion quantification, automated reporting, and reliable second opinions.

Over the past decade, Computational Linguistics (CL) and Natural Language Processing (NLP) have evolved rapidly, especially with the advent of Transformer-based Large Language Models (LLMs). This shift has transformed research goals and priorities, from Lexical and Semantic Resources to Language Modelling and Multimodality. In this study, we track the research trends of the Italian CL and NLP community through an analysis of the contributions to CLiC-it, arguably the leading Italian conference in the field. We compile the proceedings from the first 10 editions of the CLiC-it conference (from 2014 to 2024) into the CLiC-it Corpus, providing a comprehensive analysis of both its metadata, including author provenance, gender, affiliations, and more, as well as the content of the papers themselves, which address various topics. Our goal is to provide the Italian and international research communities with valuable insights into emerging trends and key developments over time, supporting informed decisions and future directions in the field.

Keyphrase extraction is a fundamental task in natural language processing. However, existing unsupervised prompt-based methods for Large Language Models (LLMs) often rely on single-stage inference pipelines with uniform prompting, regardless of document length or LLM backbone. Such one-size-fits-all designs hinder the full exploitation of LLMs' reasoning and generation capabilities, especially given the complexity of keyphrase extraction across diverse scenarios. To address these challenges, we propose MAPEX, the first framework that introduces multi-agent collaboration into keyphrase extraction. MAPEX coordinates LLM-based agents through modules for expert recruitment, candidate extraction, topic guidance, knowledge augmentation, and post-processing. A dual-path strategy dynamically adapts to document length: knowledge-driven extraction for short texts and topic-guided extraction for long texts. Extensive experiments on six benchmark datasets across three different LLMs demonstrate its strong generalization and universality, outperforming the state-of-the-art unsupervised method by 2.44% and standard LLM baselines by 4.01% in F1@5 on average. Code is available at https://github.com/NKU-LITI/MAPEX.

Mathematical reasoning poses significant challenges for Large Language Models (LLMs) due to its demand for multi-step reasoning and abstract conceptual integration. While recent test-time scaling techniques rely heavily on high-quality, challenging problems, the scarcity of Olympiad-level math problems remains a bottleneck. We introduce CogAtom, a novel cognitive atom-based framework for synthesizing mathematically rigorous and cognitively diverse problems. Unlike prior approaches, CogAtom models problem construction as a process of selecting and recombining fundamental reasoning units, cognitive atoms, extracted from human-authored solutions. A diversity-promoting random walk algorithm enables exploration of the cognitive atom space, while a constraint-based recombination mechanism ensures logical soundness and structural validity. The combinatorial nature of the graph structure provides a near-infinite space of reasoning paths, and the walk algorithm systematically explores this space to achieve large-scale synthesis of high-quality problems; meanwhile, by controlling the number of cognitive atoms, we can precisely adjust problem difficulty, ensuring diversity, scalability, and controllability of the generated problems. Experimental results demonstrate that CogAtom outperforms existing methods in accuracy, reasoning depth, and diversity, generating problems that closely match the difficulty of AIME while exceeding it in structural variation. Our work offers a cognitively grounded pathway toward scalable, high-quality math problem generation.Our code is publicly available at https://github.com/Icarus-1111/CogAtom.

Software increasingly relies on the emergent capabilities of Large Language Models (LLMs), from natural language understanding to program analysis and generation. Yet testing them on specific tasks remains difficult and costly: many prompts lack ground truth, forcing reliance on human judgment, while existing uncertainty and adequacy measures typically require full inference. A key challenge is to assess input adequacy in a way that reflects the demands of the task, ideally before even generating any output. We introduce CLOTHO, a task-specific, pre-generation adequacy measure that estimates input difficulty directly from hidden LLM states. Given a large pool of unlabelled inputs for a specific task, CLOTHO uses a Gaussian Mixture Model (GMM) to adaptively sample the most informative cases for human labelling. Based on this reference set the GMM can then rank unseen inputs by their likelihood of failure. In our empirical evaluation across eight benchmark tasks and three open-weight LLMs, CLOTHO can predict failures with a ROC-AUC of 0.716, after labelling reference sets that are on average only 5.4% of inputs. It does so without generating any outputs, thereby reducing costs compared to existing uncertainty measures. Comparison of CLOTHO and post-generation uncertainty measures shows that the two approaches complement each other. Crucially, we show that adequacy scores learnt from open-weight LLMs transfer effectively to proprietary models, extending the applicability of the approach. When prioritising test inputs for proprietary models, CLOTHO increases the average number of failing inputs from 18.7 to 42.5 out of 100, compared to random prioritisation.

The rapid spread of information via social media has made text-based fake news detection critically important due to its societal impact. This paper presents a novel detection method called Dynamic Representation and Ensemble Selection (DRES) for identifying fake news based solely on text. DRES leverages instance hardness measures to estimate the classification difficulty for each news article across multiple textual feature representations. By dynamically selecting the textual representation and the most competent ensemble of classifiers for each instance, DRES significantly enhances prediction accuracy. Extensive experiments show that DRES achieves notable improvements over state-of-the-art methods, confirming the effectiveness of representation selection based on instance hardness and dynamic ensemble selection in boosting performance. Codes and data are available at: https://github.com/FFarhangian/FakeNewsDetection_DRES