Process Reward Models (PRMs), which provide step-level feedback on reasoning traces generated by Large Language Models (LLMs), are receiving increasing attention. However, two key research gaps remain: creating PRM training data requires costly human annotation to label accurate step-level errors, and existing PRMs are limited to math reasoning domains. In response to these gaps, this paper aims to enable automatic synthesis of accurate PRM training data and the generalization of PRMs to diverse reasoning tasks beyond math reasoning. We propose FoVer, an approach to synthesize PRM training data with accurate step-level error labels automatically annotated by formal verification tools, such as Z3 and Isabelle. To show the practical effectiveness of FoVer, we synthesize a training dataset by annotating step-level error labels on LLM responses to formal logic and theorem proving tasks, without relying on human annotation. While FoVer creates training data with symbolic tasks compatible with formal verification, our experiments show that PRMs trained on our dataset exhibit cross-task generalization, enabling a single PRM to effectively perform verification across diverse reasoning tasks. Specifically, LLM-based PRMs trained with FoVer significantly outperform PRMs based on the original LLMs and achieve competitive or superior results compared to state-of-the-art PRMs, as measured by step-level verification on ProcessBench and Best-of-K performance across 12 reasoning benchmarks, including MATH, AIME, ANLI, MMLU, and BBH. The dataset and code are in the supplementary material and will be made public. The datasets, models, and code are provided at https://github.com/psunlpgroup/FoVer.

With recent breakthroughs in large-scale modeling, the Segment Anything Model (SAM) has demonstrated significant potential in a variety of visual applications. However, due to the lack of underwater domain expertise, SAM and its variants face performance limitations in end-to-end underwater instance segmentation tasks, while their higher computational requirements further hinder their application in underwater scenarios. To address this challenge, we propose a large-scale underwater instance segmentation dataset, UIIS10K, which includes 10,048 images with pixel-level annotations for 10 categories. Then, we introduce UWSAM, an efficient model designed for automatic and accurate segmentation of underwater instances. UWSAM efficiently distills knowledge from the SAM ViT-Huge image encoder into the smaller ViT-Small image encoder via the Mask GAT-based Underwater Knowledge Distillation (MG-UKD) method for effective visual representation learning. Furthermore, we design an End-to-end Underwater Prompt Generator (EUPG) for UWSAM, which automatically generates underwater prompts instead of explicitly providing foreground points or boxes as prompts, thus enabling the network to locate underwater instances accurately for efficient segmentation. Comprehensive experimental results show that our model is effective, achieving significant performance improvements over state-of-the-art methods on multiple underwater instance datasets. Datasets and codes are available at https://github.com/LiamLian0727/UIIS10K.

Vision-Language Models (VLMs) have shown capabilities in interpreting visual content, but their reliability in safety-critical scenarios remains insufficiently explored. We introduce VERI, a diagnostic benchmark comprising 200 synthetic images (100 contrastive pairs) and an additional 50 real-world images (25 pairs) for validation. Each emergency scene is paired with a visually similar but safe counterpart through human verification. Using a two-stage evaluation protocol (risk identification and emergency response), we assess 17 VLMs across medical emergencies, accidents, and natural disasters. Our analysis reveals an "overreaction problem": models achieve high recall (70-100%) but suffer from low precision, misclassifying 31-96% of safe situations as dangerous. Seven safe scenarios were universally misclassified by all models. This "better-safe-than-sorry" bias stems from contextual overinterpretation (88-98% of errors). Both synthetic and real-world datasets confirm these systematic patterns, challenging VLM reliability in safety-critical applications. Addressing this requires enhanced contextual reasoning in ambiguous visual situations.

Vision-Language Models (VLMs) show promise for autonomous driving, yet their struggle with hallucinations, inefficient reasoning, and limited real-world validation hinders accurate perception and robust step-by-step reasoning. To overcome this, we introduce \textbf{AgentThink}, a pioneering unified framework that integrates Chain-of-Thought (CoT) reasoning with dynamic, agent-style tool invocation for autonomous driving tasks. AgentThink's core innovations include: \textbf{(i) Structured Data Generation}, which establishes an autonomous driving tool library to automatically construct structured, self-verified reasoning data explicitly incorporating tool usage for diverse driving scenarios; \textbf{(ii) A Two-stage Training Pipeline}, employing Supervised Fine-Tuning (SFT) with Group Relative Policy Optimization (GRPO) to equip VLMs with the capability for autonomous tool invocation; and \textbf{(iii) Agent-style Tool-Usage Evaluation}, introducing a novel multi-tool assessment protocol to rigorously evaluate the model's tool invocation and utilization. Experiments on the DriveLMM-o1 benchmark demonstrate that AgentThink significantly boosts overall reasoning scores by \textbf{53.91%} and enhances answer accuracy by \textbf{33.54%}, while markedly improving reasoning quality and consistency. Furthermore, ablation studies and robust zero-shot/few-shot generalization experiments across various benchmarks underscore its powerful capabilities. These findings highlight a promising trajectory for developing trustworthy and tool-aware autonomous driving models. Code is available at https://github.com/curryqka/AgentThink.

Large Language Models (LLMs) are known to lack cultural representation and overall diversity in their generations, from expressing opinions to answering factual questions. To mitigate this problem, we propose multilingual prompting: a prompting method which generates several variations of a base prompt with added cultural and linguistic cues from several cultures, generates responses, and then combines the results. Building on evidence that LLMs have language-specific knowledge, multilingual prompting seeks to increase diversity by activating a broader range of cultural knowledge embedded in model training data. Through experiments across multiple models (GPT-4o, GPT-4o-mini, LLaMA 70B, and LLaMA 8B), we show that multilingual prompting consistently outperforms existing diversity-enhancing techniques such as high-temperature sampling, step-by-step recall, and persona prompting. Further analyses show that the benefits of multilingual prompting vary between high and low resource languages and across model sizes, and that aligning the prompting language with cultural cues reduces hallucination about culturally-specific information.

Recent advances in LLM agents have largely built on reasoning backbones like ReAct, which interleave thought and action in complex environments. However, ReAct often produces ungrounded or incoherent reasoning steps, leading to misalignment between the agent's actual state and goal. Our analysis finds that this stems from ReAct's inability to maintain consistent internal beliefs and goal alignment, causing compounding errors and hallucinations. To address this, we introduce ReflAct, a novel backbone that shifts reasoning from merely planning next actions to continuously reflecting on the agent's state relative to its goal. By explicitly grounding decisions in states and enforcing ongoing goal alignment, ReflAct dramatically improves strategic reliability. This design delivers substantial empirical gains: ReflAct surpasses ReAct by 27.7% on average, achieving a 93.3% success rate in ALFWorld. Notably, ReflAct even outperforms ReAct with added enhancement modules (e.g., Reflexion, WKM), showing that strengthening the core reasoning backbone is key to reliable agent performance.

We show that large language models (LLMs) exhibit an $\textit{internal chain-of-thought}$: they sequentially decompose and execute composite tasks layer-by-layer. Two claims ground our study: (i) distinct subtasks are learned at different network depths, and (ii) these subtasks are executed sequentially across layers. On a benchmark of 15 two-step composite tasks, we employ layer-from context-masking and propose a novel cross-task patching method, confirming (i). To examine claim (ii), we apply LogitLens to decode hidden states, revealing a consistent layerwise execution pattern. We further replicate our analysis on the real-world $\text{TRACE}$ benchmark, observing the same stepwise dynamics. Together, our results enhance LLMs transparency by showing their capacity to internally plan and execute subtasks (or instructions), opening avenues for fine-grained, instruction-level activation steering.

World models, which predict future transitions from past observation and action sequences, have shown great promise for improving data efficiency in sequential decision-making. However, existing world models often require extensive domain-specific training and still produce low-fidelity, coarse predictions, limiting their usefulness in complex environments. In contrast, video diffusion models trained on large-scale internet data have demonstrated impressive capabilities in generating high-quality videos that capture diverse real-world dynamics. In this work, we present Vid2World, a general approach for leveraging and transferring pre-trained video diffusion models into interactive world models. To bridge the gap, Vid2World systematically explores video diffusion causalization, reshaping both the architecture and training objective of pre-trained models to enable autoregressive generation. Additionally, it incorporates a causal action guidance mechanism to enhance action controllability in the resulting interactive world models. Extensive experiments across multiple domains, including robot manipulation, 3D game simulation, and open-world navigation, demonstrate that our method offers a scalable and effective pathway for repurposing highly capable video diffusion models into interactive world models.

In-context Learning (ICL) utilizes structured demonstration-query inputs to induce few-shot learning on Language Models (LMs), which are not originally pre-trained on ICL-style data. To bridge the gap between ICL and pre-training, some approaches fine-tune LMs on large ICL-style datasets by an end-to-end paradigm with massive computational costs. To reduce such costs, in this paper, we propose Attention Behavior Fine-Tuning (ABFT), utilizing the previous findings on the inner mechanism of ICL, building training objectives on the attention scores instead of the final outputs, to force the attention scores to focus on the correct label tokens presented in the context and mitigate attention scores from the wrong label tokens. Our experiments on 9 modern LMs and 8 datasets empirically find that ABFT outperforms in performance, robustness, unbiasedness, and efficiency, with only around 0.01% data cost compared to the previous methods. Moreover, our subsequent analysis finds that the end-to-end training objective contains the ABFT objective, suggesting the implicit bias of ICL-style data to the emergence of induction heads. Our work demonstrates the possibility of controlling specific module sequences within LMs to improve their behavior, opening up the future application of mechanistic interpretability.

Numerous works have noted similarities in how machine learning models represent the world, even across modalities. Although much effort has been devoted to uncovering properties and metrics on which these models align, surprisingly little work has explored causes of this similarity. To advance this line of inquiry, this work explores how two factors - dataset overlap and task overlap - influence downstream model similarity. We evaluate the effects of both factors through experiments across model sizes and modalities, from small classifiers to large language models. We find that both task and dataset overlap cause higher representational similarity and that combining them provides the strongest effect. Finally, we consider downstream consequences of representational similarity, demonstrating how greater similarity increases vulnerability to transferable adversarial and jailbreak attacks.