Large Language Models (LLMs) are predominantly evaluated on Standard American English (SAE), often overlooking the diversity of global English varieties.This narrow focus may raise fairness concerns as degraded performance on non-standard varieties can lead to unequal benefits for users worldwide.Therefore, it is critical to extensively evaluate the linguistic robustness of LLMs on multiple non-standard English varieties.We introduce Trans-EnV, a framework that automatically transforms SAE datasets into multiple English varieties to evaluate the linguistic robustness. Our framework combines (1) linguistics expert knowledge to curate variety-specific features and transformation guidelines from linguistic literature and corpora, and (2) LLM-based transformations to ensure both linguistic validity and scalability.Using Trans-EnV, we transform six benchmark datasets into 38 English varieties and evaluate seven state-of-the-art LLMs.Our results reveal significant performance disparities, with accuracy decreasing by up to 46.3% on non-standard varieties.These findings highlight the importance of comprehensive linguistic robustness evaluation across diverse English varieties. Each construction of Trans-EnV was validated through rigorous statistical testing and consultation with a researcher in the field of second language acquisition, ensuring its linguistic validity.Our code and datasets are publicly available.

We study the use of Process Reward Models (PRMs) for guiding Long Chain-of-Thought (CoT) reasoning in large language models. Although PRMs deliver fine-grained feedback in standard tasks, PRM-guided beam search does not consistently outperform PRM-free approaches in long CoT reasoning. We trace this shortfall to a step quality degradation''--the expected step quality shows concave behavior, yielding unimodal or monotonically declining trends. To counteract this, we propose Z-Score Guided Early Stopping (ZGES), which halts search at the detected quality peak using local PRM-reward z-scores. Across multiple math benchmarks and model scales, ZGES outperforms both standard PRM-guided beam search and the PRM-free methods.

Achieving Artificial General Intelligence (AGI) requires AI agents that can not only make strategic decisions but also engage in flexible and meaningful communication. Inspired by Wittgenstein's language game theory, we propose that language agents can learn through in-context interaction rather than traditional multi-stage frameworks that separate decision-making from language expression. Using Werewolf, a social deduction game that tests language understanding, strategic interaction, and adaptability, as a test bed, we develop the Multi-agent Kahneman-Tversky's Optimization (MaKTO). MaKTO engages diverse models in extensive gameplay to generate unpaired desirable and unacceptable responses, then employs KTO to refine the model's decision-making process. In 9-player Werewolf games, MaKTO achieves a 61% average win rate across various models, outperforming GPT-4o and two-stage RL agents by relative improvements of 23.0% and 10.9%, respectively. Notably, MaKTO also demonstrates human-like performance, winning 60% against expert players and showing only 48.9% detectability in Turing-style blind tests.

Large language models (LLMs) have achieved remarkable progress in natural language generation, yet they continue to display puzzling behaviors--such as repetition and incoherence--even when exhibiting low perplexity. This highlights a key limitation of conventional evaluation metrics, which emphasize local prediction accuracy while overlooking long-range structural complexity. We introduce correlation dimension, a fractal-geometric measure of self-similarity, to quantify the epistemological complexity of text as perceived by a language model.

It was published by KPMG, one of the world's'Big Four' accounting firms. In October last year, KPMG published a report titled, which was about how companies are using AI to cater to customers' needs. KPMG is one of the Big Four professional services and accounting firms in the world, along with Deloitte, PricewaterhouseCoopers and Ernst & Young. Apparently, though, that report was full of AI hallucinations and included examples of agentic AIs that either did not exist or did not have the capabilities KPMG stated in the paper. Investigators for GPTZero, the maker of an AI content detection tool, found inaccuracies and fake footnotes all over the report, which were also verified by the Financial Times .

Inspired by recent findings on the fractal geometry of language, we introduce Recursive INference Scaling (RINS) as a complementary, plug-in recipe for scaling inference time in language and multimodal systems. RINS is a particular form of recursive depth that significantly outperforms +55 other variants, including the recent repeat-all-over (RAO) strategy in Mobile LLM (Liu et al., 2024) and latent recurrent thinking (Geiping et al., 2025). Unlike prior works, we carry out our comparisons on a compute-matched regime, and demonstrate that for a fixed model size and training compute budget, RINS substantially improves language modeling performance.

Source-free domain adaptation (SFDA) is a challenging task that tackles domain shifts using only a pre-trained source model and unlabeled target data. Existing SFDA methods are restricted by the fundamental limitation of source-target domain discrepancy. Non-generation SFDA methods suffer from unreliable pseudo-labels in challenging scenarios with large domain discrepancies, while generation-based SFDA methods are evidently degraded due to enlarged domain discrepancies in creating pseudo-source data. To address this limitation, we propose a novel generation-based framework named Diffusion-Driven Progressive Target Manipulation (DPTM) that leverages unlabeled target data as references to reliably generate and progressively refine a pseudo-target domain for SFDA. Specifically, we divide the target samples into a trust set and a non-trust set based on the reliability of pseudo-labels to sufficiently and reliably exploit their information. For samples from the non-trust set, we develop a manipulation strategy to semantically transform them into the newly assigned categories, while simultaneously maintaining them in the target distribution via a latent diffusion model. Furthermore, we design a progressive refinement mechanism that progressively reduces the domain discrepancy between the pseudo-target domain and the real target domain via iterative refinement. Experimental results demonstrate that DPTM outperforms existing methods by a large margin and achieves state-of-the-art performance on four prevailing SFDA benchmark datasets with different scales. Remarkably, DPTM can significantly enhance the performance by up to 18.6\% in scenarios with large source-target gaps.

Many decision-making tasks, where both accuracy and efficiency matter, still require human supervision. For example, tasks like traffic officers reviewing hour-long dashcam footage or researchers screening conference videos can benefit from concise summaries that reduce cognitive load and save time. Yet current vision-language models (VLMs) often produce verbose, redundant outputs that hinder task performance. Existing video caption evaluation depends on costly human annotations and overlooks the summaries' utility in downstream tasks. We address these gaps with $\underline{\textbf{V}}$ideo-to-text $\underline{\textbf{I}}$nformation $\underline{\textbf{B}}$ottleneck $\underline{\textbf{E}}$valuation (VIBE), an annotation-free method that scores VLM outputs using two metrics: $\textit{grounding}$ (how well the summary aligns with visual content) and $\textit{utility}$ (how informative it is for the task). VIBE selects from randomly sampled VLM outputs by ranking them according to the two scores to support effective human decision-making. Human studies on $\texttt{LearningPaper24}$, $\texttt{SUTD-TrafficQA}$, and $\texttt{LongVideoBench}$ show that summaries selected by VIBE consistently improve performance--boosting task accuracy by up to $61.23$% and reducing response time by $75.77$% compared to naive VLM summaries or raw video.

Reinforcement learning (RL) heavily depends on well-designed reward functions, which are often biased and difficult to design for complex behaviors. Preference-based RL (PbRL) addresses this by learning reward models from human feedback, but its practicality is constrained by a critical dilemma: while existing methods reduce human effort through query optimization, they neglect the preference buffer's restricted coverage -- a factor that fundamentally determines the reliability of reward model. We systematically demonstrate this limitation creates distributional mismatch: reward models trained on static buffers reliably assess in-distribution trajectories but falter with out-of-distribution (OOD) trajectories from policy exploration.

Prediction-Powered Inference (PPI) is a recently proposed statistical inference technique for parameter estimation that leverages pseudo-labels on both labeled and unlabeled data to construct an unbiased, low-variance estimator. In this work, we extend its core idea to semi-supervised learning (SSL) for model training, introducing a novel unbiased gradient estimator. This extension addresses a key challenge in SSL: while unlabeled data can improve model performance, its benefit heavily depends on the quality of pseudo-labels. Inaccurate pseudo-labels can introduce bias, leading to suboptimal models. To balance the contributions of labeled and pseudo-labeled data, we utilize an interpolation parameter and tune it on the fly, alongside the model parameters, using a one-dimensional online learning algorithm. We verify the practical advantage of our approach through experiments on both synthetic and real datasets, demonstrating improved performance over classic SSL baselines and PPI methods that tune the interpolation parameter offline.