We introduce OpenFActScore, an open-source implementation of the FActScore framework for evaluating the factuality of text generated by large language models (LLMs). FActScore evaluates the factual accuracy of long-form text by using Atomic Fact Generation (AFG) to extract individual factual claims and Atomic Fact Validation (AFV) to verify each claim against a trusted knowledge source. While the original FActScore relies on closed-source and commercial models such as InstructGPT and ChatGPT, OpenFActScore enables the use of any Hugging Face-compatible model for both AFG and AFV. We provide a detailed technical overview of our implementation, highlighting design choices and modifications made to support open models. We evaluate multiple open-source LLMs on both AFG and AFV using the original FActScore benchmark, reporting BERTScore-F1 for AFG and Error Rate relative to human annotations for AFV. Our results show that open models can approximate the performance of closed-source systems, with Gemma achieving the best overall performance, and our final setup obtains a 0.99 Pearson correlation with the original FActScore experiments. OpenFActScore promotes transparency, reproducibility, and cost-effective evaluation, and is available at: https://github.com/lflage/OpenFActScore.

Translating natural language mathematical statements into formal, executable code is a fundamental challenge in automated theorem proving. While prior work has focused on generation and compilation success, little attention has been paid to the critic phase-the evaluation of whether generated formalizations truly capture the semantic intent of the original problem. In this paper, we introduce CriticLean, a novel critic-guided reinforcement learning framework that elevates the role of the critic from a passive validator to an active learning component. Specifically, first, we propose the CriticLeanGPT, trained via supervised fine-tuning and reinforcement learning, to rigorously assess the semantic fidelity of Lean 4 formalizations. Then, we introduce CriticLeanBench, a benchmark designed to measure models' ability to distinguish semantically correct from incorrect formalizations, and demonstrate that our trained CriticLeanGPT models can significantly outperform strong open- and closed-source baselines. Building on the CriticLean framework, we construct FineLeanCorpus, a dataset comprising over 285K problems that exhibits rich domain diversity, broad difficulty coverage, and high correctness based on human evaluation. Overall, our findings highlight that optimizing the critic phase is essential for producing reliable formalizations, and we hope our CriticLean will provide valuable insights for future advances in formal mathematical reasoning.

This paper presents our submission to Task 1, Subjectivity Detection, of the CheckThat! Lab at CLEF 2025. We investigate the effectiveness of transfer-learning and stylistic data augmentation to improve classification of subjective and objective sentences in English news text. Our approach contrasts fine-tuning of pre-trained encoders and transfer-learning of fine-tuned transformer on related tasks. We also introduce a controlled augmentation pipeline using GPT-4o to generate paraphrases in predefined subjectivity styles. To ensure label and style consistency, we employ the same model to correct and refine the generated samples. Results show that transfer-learning of specified encoders outperforms fine-tuning general-purpose ones, and that carefully curated augmentation significantly enhances model robustness, especially in detecting subjective content. Our official submission placed us $16^{th}$ of 24 participants. Overall, our findings underscore the value of combining encoder specialization with label-consistent augmentation for improved subjectivity detection. Our code is available at https://github.com/dsgt-arc/checkthat-2025-subject.

Large reasoning models (LRMs) have demonstrated impressive capabilities in domains like mathematics and program synthesis. Despite their strong performance, LRMs often exhibit overthinking -- excessive and redundant reasoning steps that introduce inefficiencies during inference. This phenomenon raises an important question for LRM self-evaluation: How can a model autonomously assess the correctness of its own reasoning trajectory without external labels? To address this, we propose Chain-of-Reasoning Embedding (CoRE), a series of hidden states in latent space to enable label-free self-evaluation on intermediate reasoning steps of LRMs, so as to enhance metacognition abilities for improved reasoning efficiency. By analyzing the geometric properties of the CoRE trajectories, we reveal that redundant reasoning usually presents cyclical fluctuations, which correspond to repetitive and unconscious reflection/exploration. Leveraging this insight, we further introduce a training-free, label-free self-evaluation framework, CoRE-Eval, to detect such patterns and dynamically determine whether to terminate reasoning early. Extensive experiments on mathematical reasoning benchmarks (GSM8K, MATH-500, and AIME) and across model sizes from 7B to 32B demonstrate that CoRE-Eval reduces chain-of-thought length by 13.7% to 33.2% while improving answer accuracy by around 10%, achieving 70.0% accuracy on the challenging AIME benchmark with the 32B model.

This paper presents Auto-RubikAI, a modular autonomous planning framework that integrates a symbolic Knowledge Base (KB), a vision-language model (VLM), and a large language model (LLM) to solve structured manipulation tasks exemplified by Rubik's Cube restoration. Unlike traditional robot systems based on predefined scripts, or modern approaches relying on pretrained networks and large-scale demonstration data, Auto-RubikAI enables interpretable, multi-step task execution with minimal data requirements and no prior demonstrations. The proposed system employs a KB module to solve group-theoretic restoration steps, overcoming LLMs' limitations in symbolic reasoning. A VLM parses RGB-D input to construct a semantic 3D scene representation, while the LLM generates structured robotic control code via prompt chaining. This tri-module architecture enables robust performance under spatial uncertainty. We deploy Auto-RubikAI in both simulation and real-world settings using a 7-DOF robotic arm, demonstrating effective Sim-to-Real adaptation without retraining. Experiments show a 79% end-to-end task success rate across randomized configurations. Compared to CFOP, DeepCubeA, and Two-Phase baselines, our KB-enhanced method reduces average solution steps while maintaining interpretability and safety. Auto-RubikAI provides a cost-efficient, modular foundation for embodied task planning in smart manufacturing, robotics education, and autonomous execution scenarios. Code, prompts, and hardware modules will be released upon publication.

The widespread deployment of large language models (LLMs) across critical domains has amplified the societal risks posed by algorithmically generated misinformation. Unlike traditional false content, LLM-generated misinformation can be self-reinforcing, highly plausible, and capable of rapid propagation across multiple languages, which traditional detection methods fail to mitigate effectively. This paper introduces a proactive defense paradigm, shifting from passive post hoc detection to anticipatory mitigation strategies. We propose a Three Pillars framework: (1) Knowledge Credibility, fortifying the integrity of training and deployed data; (2) Inference Reliability, embedding self-corrective mechanisms during reasoning; and (3) Input Robustness, enhancing the resilience of model interfaces against adversarial attacks. Through a comprehensive survey of existing techniques and a comparative meta-analysis, we demonstrate that proactive defense strategies offer up to 63\% improvement over conventional methods in misinformation prevention, despite non-trivial computational overhead and generalization challenges. We argue that future research should focus on co-designing robust knowledge foundations, reasoning certification, and attack-resistant interfaces to ensure LLMs can effectively counter misinformation across varied domains.

Measuring the generalization ability of Large Language Models (LLMs) is challenging due to data contamination. As models grow and computation becomes cheaper, ensuring tasks and test cases are unseen during training phases will become nearly impossible. We argue that knowledge-retrieval and reasoning tasks are not ideal for measuring generalization, as LLMs are not trained for specific tasks. Instead, we propose user behavior prediction, also a key aspect of personalization, as a theoretically sound, scalable, and robust alternative. We introduce a novel framework for this approach and test it on movie and music recommendation datasets for GPT-4o, GPT-4o-mini, and Llama-3.1-8B-Instruct. Results align with our framework's predictions, showing GPT-4o outperforms GPT-4o-mini and Llama, though all models have much room for improvement, especially Llama.

Large language models (LLMs) trained on text demonstrated remarkable results on natural language processing (NLP) tasks. These models have been adapted to decipher the language of DNA, where sequences of nucleotides act as "words" that encode genomic functions. However, the genome differs fundamentally from natural language, as it lacks clearly defined words or a consistent grammar. Although DNA language models (DNALMs) such as DNABERT, GENA-LM have achieved high level of performance on genome-related biological tasks, these models do not encode biological functions in the presence of sequence variations. To address this problem, we pre-train foundation models that effectively integrate sequence variations, in particular Single Nucleotide Polymorphisms (SNPs), as they underlie important biological functions. Specifically, we use ModernBERT to pre-train two different Biomedical Foundation Models (BMFM), namely, BMFM-DNA-REF in which the model is trained with sequences of varying lengths along with their reverse complements derived from the reference genome and BMFM-DNA-SNP in which the model is trained with sequences created using a novel representation scheme that encodes sequence variations. Our findings indicate that integrating sequence variations into DNALMs helps capture the biological functions as seen in improvements on all fine-tuning tasks. To explore the model's practical utility, we experimented with various strategies for SNP imputation on promoter detection task introduced in DNABERT-2. However, we acknowledge that the current benchmarks are limited in their ability to fully evaluate these models. To enable more comprehensive assessment in the future and encourage community contributions, we release our models through HuggingFace and the code to reproduce the results at https://github.com/BiomedSciAI/biomed-multi-omic

The rapid advancements of AI agents have ignited the long-held ambition of leveraging them to accelerate scientific discovery. Achieving this goal requires a deep understanding of the frontiers of human knowledge. As such, Humanity's Last Exam (HLE) provides an exceptionally challenging touchstone for evaluating scientific AI agents. In this work, we aim to construct the foundational architecture for general-purpose agents and validate the capabilities through leading performance on HLE. To achieve this, we introduce X-Master, a tool-augmented reasoning agent designed to emulate human researchers by interacting flexibly with external tools during its reasoning process. This agent, guided by the conceptualization of code as an interaction language, can flexibly leverage built-in Python libraries and our customized tools to augment the reasoning. We further scale its capabilities through X-Masters, a scattered-and-stacked agentic workflow that systematically enhances breadth and depth of reasoning. Our open-source solution, X-Masters, sets a new state-of-the-art record on HLE with a score of 32.1%, surpassing OpenAI's and Google's Deep Research (26.6% and 26.9%) and becoming the first to exceed the 30% threshold. This work allows us to gain a deeper understanding of complex task-solving and accumulates valuable experience that can inform future advancements, guiding subsequent model training.

Pre-trained language models (PLMs) are widely used to derive semantic representations from item metadata in recommendation and search. In sequential recommendation, PLMs enhance ID-based embeddings through textual metadata, while in product search, they align item characteristics with user intent. Recent studies suggest task and domain-specific fine-tuning are needed to improve representational power. This paper challenges this assumption, showing that Generalist Text Embedding Models (GTEs), pre-trained on large-scale corpora, can guarantee strong zero-shot performance without specialized adaptation. Our experiments demonstrate that GTEs outperform traditional and fine-tuned models in both sequential recommendation and product search. We attribute this to a superior representational power, as they distribute features more evenly across the embedding space. Finally, we show that compressing embedding dimensions by focusing on the most informative directions (e.g., via PCA) effectively reduces noise and improves the performance of specialized models. To ensure reproducibility, we provide our repository at https://split.to/gte4ps.