Agents built on vision-language models increasingly face tasks that demand anticipating future states rather than relying on short-horizon reasoning. Generative world models offer a promising remedy: agents could use them as external simulators to foresee outcomes before acting. This paper empirically examines whether current agents can leverage such world models as tools to enhance their cognition. Across diverse agentic and visual question answering tasks, we observe that some agents rarely invoke simulation (fewer than 1%), frequently misuse predicted rollouts (approximately 15%), and often exhibit inconsistent or even degraded performance (up to 5%) when simulation is available or enforced. Attribution analysis further indicates that the primary bottleneck lies in the agents’ capacity to decide when to simulate, how to interpret predicted outcomes, and how to integrate foresight into downstream reasoning. These findings underscore the need for mechanisms that foster calibrated, strategic interaction with world models, paving the way toward more reliable anticipatory cognition in future agent systems.

Video-LLMs face a fundamental tension in long-video reasoning: static, sparse frame sampling either dilutes evidence across task-irrelevant segments at significant cost or misses fine-grained temporal semantics altogether. We propose a novel, cognitively-inspired task – Endomorphic Multimodal Compression (EMC) – as a structurally-constrained sufficient-statistic problem for VideoQA, and formulate it as an endomorphic transformation F_EMC : (V, Q) -> (v, q) that compresses the multimodal input while preserving answer invariance across reasonable downstream models. The endomorphic form keeps the compressed output in the downstream pipeline's native task space – a structural mirror of the filter-then-reason mechanism in the cognitive literature motivating EMC – distinguishing it from latent-code compression (IB / VIB) and making the formulation extensible to other multimodal settings. Under the Markov chain A -> (V, Q) -> (v, q), EMC realizes the classical sufficiency condition I((v, q); A) = I((V, Q); A) in its VideoQA-natural form. As a modular front-end, EMC plugs into both Video Instruction Tuning and Video Question Answering pipelines. We release the first dedicated benchmark and propose ReSimplifyIt, an EMC baseline surpassing prior methods by 0.40 F-1 with competitive query rewriting. Integrating EMC yields relative gains of 7.33

Reasoning is a fundamental cognitive process underlying inference, problem-solving, and decision-making. While large language models (LLMs) demonstrate strong reasoning capabilities in closed-world settings, they struggle in open-ended and dynamic environments. Agentic reasoning marks a paradigm shift by reframing LLMs as autonomous agents that plan, act, and learn through continual interaction. In this survey, we organize agentic reasoning along three complementary dimensions. First, we characterize environmental dynamics through three layers: foundational agentic reasoning, which establishes core single-agent capabilities including planning, tool use, and search in stable environments; self-evolving agentic reasoning, which studies how agents refine these capabilities through feedback, memory, and adaptation; and collective multi-agent reasoning, which extends intelligence to collaborative settings involving coordination, knowledge sharing, and shared goals. Across these layers, we distinguish in-context reasoning, which scales test-time interaction through structured orchestration, from post-training reasoning, which optimizes behaviors via reinforcement learning and supervised fine-tuning. We further review representative agentic reasoning frameworks across real-world applications and benchmarks, including science, robotics, healthcare, autonomous research, and mathematics. This survey synthesizes agentic reasoning methods into a unified roadmap bridging thought and action, and outlines open challenges and future directions, including personalization, long-horizon interaction, world modeling, scalable multi-agent training, and governance for real-world deployment.

Memory-augmented large language models extend reasoning beyond a fixed context window by maintaining long-term memory across interactions. However, existing memory systems often collapse stable user facts, episodic events, and behavioral rules into a shared space, allowing functionally distinct memories to be retrieved and used as interchangeable evidence. We identify this failure mode as heterogeneous memory contamination, where context-specific events become overgeneralized claims, or semantically relevant but functionally incompatible memories mislead generation. To this end, we introduce MemGuard, a type-aware memory framework that preserves functional memory boundaries during memory construction and retrieval. It assigns each memory an explicit functional role at write time, maintains relations across type-isolated memories, and selectively composes evidence only from necessary memory types, reducing contamination from irrelevant or functionally incompatible evidence. Across hallucination and long-horizon conversation benchmarks, MemGuard improves memory reliability by up to 28.27

Thematic analysis (TA) is a widely used qualitative approach for uncovering latent meanings in unstructured text data. TA provides valuable insights in healthcare but is resource-intensive. Large Language Models (LLMs) have been introduced to perform TA, yet their applications in high-stakes healthcare settings, particularly for qualitative clinical interview analysis, remain limited. Here, we propose TAMA: A Human-AI Collaborative Thematic Analysis framework using Multi-Agent LLMs for clinical interviews. We leverage the scalability and coherence of multi-agent systems through structured conversations between agents and coordinate the expertise of cardiac experts in TA. Using interview transcripts from parents of children with Anomalous Aortic Origin of a Coronary Artery (AAOCA), a rare congenital heart disease, we demonstrate that TAMA outperforms single-agent LLM TA approaches, achieving higher thematic hit rate, coverage, and distinctiveness. TAMA demonstrates strong potential for automated TA in clinical settings by leveraging multi-agent LLM systems with human-in-the-loop integration by enhancing quality while significantly reducing manual workload. The full implementation is publicly available at https://github.com/Charlie-Yi-SJ/TAMA.

Scientific texts often convey authority due to their technical language and complex data. However, this complexity can sometimes lead to the spread of misinformation. Non-experts are particularly susceptible to misleading claims based on scientific tables due to their high information density and perceived credibility. Existing table claim verification models, including state-of-the-art large language models (LLMs), often struggle with precise fine-grained reasoning, resulting in errors and a lack of precision in verifying scientific claims. Inspired by Cognitive Load Theory, we propose that enhancing a model's ability to interpret table-based claims involves reducing cognitive load by developing modular, reusable reasoning components (i.e., atomic skills). We introduce a skill-chaining schema that dynamically composes these skills to facilitate more accurate and generalizable reasoning with a reduced cognitive load. To evaluate this, we create SciAtomicBench, a cross-domain benchmark with fine-grained reasoning annotations. With only 350 fine-tuning examples, our model trained by atomic reasoning outperforms GPT-4o's chain-of-thought method, achieving state-of-the-art results with far less training data.

Self-consistency-based approaches, which involve repeatedly sampling multiple outputs and selecting the most consistent one as the final response, prove to be remarkably effective in improving the factual accuracy of large language models. Nonetheless, existing methods usually have strict constraints on the task format, largely limiting their applicability. In this paper, we present Integrative Decoding (ID), to unlock the potential of self-consistency in open-ended generation tasks. ID operates by constructing a set of inputs, each prepended with a previously sampled response, and then processes them concurrently, with the next token being selected by aggregating of all their corresponding predictions at each decoding step. In essence, this simple approach implicitly incorporates self-consistency in the decoding objective. Extensive evaluation shows that ID consistently enhances factuality over a wide range of language models, with substantial improvements on the TruthfulQA (+11.2%), Biographies (+15.4%) and LongFact (+8.5%) benchmarks. The performance gains amplify progressively as the number of sampled responses increases, indicating the potential of ID to scale up with repeated sampling.

Hallucination is a persistent challenge in large language models (LLMs), where even with rigorous quality control, models often generate distorted facts. This paradox, in which error generation continues despite high-quality training data, calls for a deeper understanding of the underlying LLM mechanisms. To address it, we propose a novel concept: knowledge overshadowing, where model's dominant knowledge can obscure less prominent knowledge during text generation, causing the model to fabricate inaccurate details. Building on this idea, we introduce a novel framework to quantify hallucinations by modeling knowledge overshadowing. Central to our approach is the log-linear law, which predicts that the rate of hallucination increases linearly with the logarithmic scale of (1) Knowledge Popularity, (2) Knowledge Length, and (3) Model Size. The law provides a means to preemptively quantify hallucinations, offering foresight into their occurrence even before model training or inference. Built on the overshadowing effect, we propose a new decoding strategy CoDA, to mitigate hallucinations, which notably enhances model factuality on Overshadow (27.9%), MemoTrap (13.1%) and NQSwap (18.3%). Our findings not only deepen understandings of the underlying mechanisms behind hallucinations but also provide actionable insights for developing more predictable and controllable language models.

Recent progress in large language models (LLMs) has enabled substantial advances in solving mathematical problems. However, existing benchmarks often fail to reflect the complexity of real-world problems, which demand open-ended, interdisciplinary reasoning and integration of computational tools. To address this gap, we introduce ModelingBench, a novel benchmark featuring real-world-inspired, open-ended problems from math modeling competitions across diverse domains, ranging from urban traffic optimization to ecosystem resource planning. These tasks require translating natural language into formal mathematical formulations, applying appropriate tools, and producing structured, defensible reports. ModelingBench also supports multiple valid solutions, capturing the ambiguity and creativity of practical modeling. We also present ModelingAgent, a multi-agent framework that coordinates tool use, supports structured workflows, and enables iterative self-refinement to generate well-grounded, creative solutions. To evaluate outputs, we further propose ModelingJudge, an expert-in-the-loop system leveraging LLMs as domain-specialized judges assessing solutions from multiple expert perspectives. Empirical results show that ModelingAgent substantially outperforms strong baselines and often produces solutions indistinguishable from those of human experts. Together, our work provides a comprehensive framework for evaluating and advancing real-world problem-solving in open-ended, interdisciplinary modeling challenges.

The dynamic nature of real-world information necessitates efficient knowledge editing (KE) in large language models (LLMs) for knowledge updating. However, current KE approaches, which typically operate on (subject, relation, object) triples, ignore the contextual information and the relation among different knowledge. Such editing methods could thus encounter an uncertain editing boundary, leaving a lot of relevant knowledge in ambiguity: Queries that could be answered pre-edit cannot be reliably answered afterward. In this work, we analyze this issue by introducing a theoretical framework for KE that highlights an overlooked set of knowledge that remains unchanged and aids in knowledge deduction during editing, which we name as the deduction anchor. We further address this issue by proposing a novel task of event-based knowledge editing that pairs facts with event descriptions. This task manifests not only a closer simulation of real-world editing scenarios but also a more logically sound setting, implicitly defining the deduction anchor to address the issue of indeterminate editing boundaries. We empirically demonstrate the superiority of event-based editing over the existing setting on resolving uncertainty in edited models, and curate a new benchmark dataset EvEdit derived from the CounterFact dataset. Moreover, while we observe that the event-based setting is significantly challenging for existing approaches, we propose a novel approach Self-Edit that showcases stronger performance, achieving 55.6% consistency improvement while maintaining the naturalness of generation.

Hallucination is often regarded as a major impediment for using large language models (LLMs), especially for knowledge-intensive tasks. Even when the training corpus consists solely of true statements, language models still generate hallucinations in the form of amalgamations of multiple facts. We coin this phenomenon as “knowledge overshadowing”: when we query knowledge from a language model with multiple conditions, some conditions overshadow others, leading to hallucinated outputs. This phenomenon partially stems from training data imbalance, which we verify on both pretrained models and fine-tuned models, over a wide range of LM model families and sizes.From a theoretical point of view, knowledge overshadowing can be interpreted as over-generalization of the dominant conditions (patterns). We show that the hallucination rate grows with both the imbalance ratio (between the popular and unpopular condition) and the length of dominant condition description, consistent with our derived generalization bound. Finally, we propose to utilize overshadowing conditions as a signal to catch hallucination before it is produced, along with a training-free self-contrastive decoding method to alleviate hallucination during inference. Our proposed approach showcases up to 82 with different models and datasets.

Honesty is a fundamental principle for aligning large language models (LLMs) with human values, requiring these models to recognize what they know and don't know and be able to faithfully express their knowledge. Despite promising, current LLMs still exhibit significant dishonest behaviors, such as confidently presenting wrong answers or failing to express what they know. In addition, research on the honesty of LLMs also faces challenges, including varying definitions of honesty, difficulties in distinguishing between known and unknown knowledge, and a lack of comprehensive understanding of related research. To address these issues, we provide a survey on the honesty of LLMs, covering its clarification, evaluation approaches, and strategies for improvement. Moreover, we offer insights for future research, aiming to inspire further exploration in this important area.

The dynamic nature of real-world information necessitates knowledge editing (KE) in large language models (LLMs). The edited knowledge should propagate and facilitate the deduction of new information based on existing model knowledge. We term the existing related knowledge in LLM serving as the origination of knowledge propagation as ”deduction anchors”. However, current KE approaches, which only operate on (subject, relation, object) triple. We both theoretically and empirically observe that this simplified setting often leads to uncertainty when determining the deduction anchors, causing low confidence in their answers. To mitigate this issue, we propose a novel task of event-based knowledge editing that pairs facts with event descriptions. This task manifests not only a closer simulation of real-world editing scenarios but also a more logically sound setting, implicitly defining the deduction anchor and enabling LLMs to propagate knowledge confidently. We curate a new benchmark dataset Evedit derived from the CounterFact dataset and validate its superiority in improving model confidence. Moreover, while we observe that the event-based setting is significantly challenging for existing approaches, we propose a novel approach Self-Edit that showcases stronger performance, achieving 55.6% consistency improvement while maintaining the naturalness of generation.

Discrimination against minority groups within the banking sector has long resulted in unequal treatment in financial services. Recent works in the general machine learning domain can promote group fairness for predictions on static tabular data, but their direct application in finance often proves ineffective. Financial losses of banks may arise from inaccurate predictions due to the overlooked dynamic nature of data, and illegal discrimination against some individual clients could still occur since fairness is promoted on the subgroup level. Therefore, we model the data as a dynamic or temporal transaction network for better utility and investigate individual fairness on this dynamic graph for the loan approval task. We define two novel individual fairness properties on temporal graphs with a theoretical analysis of their respective regret. Using these notions, we design a temporally fair graph neural network (TF-GNN) approach under a new real-time evaluation scheme for dynamic transaction networks. Experiments on real-world datasets demonstrate the superiority of the proposed method for both utility improvement in accuracy and fairness promotion in NDCG@k.

Language features are evolving in real-world social media, resulting in the deteriorating performance of text classification in dynamics. To address this challenge, we study temporal adaptation, where models trained on past data are tested in the future. Most prior work focused on continued pretraining or knowledge updating, which may compromise their performance on noisy social media data. To tackle this issue, we reflect feature change via modeling latent topic evolution and propose a novel model, VIBE: Variational Information Bottleneck for Evolutions. Concretely, we first employ two Information Bottleneck (IB) regularizers to distinguish past and future topics. Then, the distinguished topics work as adaptive features via multi-task training with timestamp and class label prediction. In adaptive learning, VIBE utilizes retrieved unlabeled data from online streams created posterior to training data time. Substantial Twitter experiments on three classification tasks show that our model, with only 3% of data, significantly outperforms previous state-of-the-art continued-pretraining methods.

Language features are ever-evolving in the real-world social media environment. Many trained models in natural language understanding (NLU), ineffective in semantic inference for unseen features, might consequently struggle with the deteriorating performance in dynamicity. To address this challenge, we empirically study social media NLU in a dynamic setup, where models are trained on the past data and test on the future. It better reflects the realistic practice compared to the commonly-adopted static setup of random data split. To further analyze model adaption to the dynamicity, we explore the usefulness of leveraging some unlabeled data created after a model is trained. The performance of unsupervised domain adaption baselines based on auto-encoding and pseudo-labeling and a joint framework coupling them both are examined in the experiments. Substantial results on four social media tasks imply the universally negative effects of evolving environments over classification accuracy, while auto-encoding and pseudo-labeling collaboratively show the best robustness in dynamicity.

Millions of hashtags are created on social media every day to cross-refer messages concerning similar topics. To help people find the topics they want to discuss, this paper characterizes a user's hashtagging preferences via predicting how likely they will post with a hashtag. It is hypothesized that one's interests in a hashtag are related to what they said before (user history) and the existing posts present the hashtag (hashtag contexts). These factors are married in the deep semantic space built with a pre-trained BERT and a neural topic model via joint training. In this way, user interests learned from the past can be customized to match future hashtags, which is beyond the capability of existing methods assuming unchanged hashtag semantics. Furthermore, we propose a novel personalized topic attention to capture salient contents to personalize hashtag contexts. Experiments on a large-scale Twitter dataset show that our model significantly outperforms the state-of-the-art recommendation approach without exploiting latent topics.(1)

This paper presents a novel task to generate poll questions for social media posts. It offers an easy way to hear the voice from the public and learn from their feelings to important social topics. While most related work tackles formally-written texts (e.g., exam papers), we generate poll questions for short and colloquial social media messages exhibiting severe data sparsity. To deal with that, we propose to encode user comments and discover latent topics therein as contexts. They are then incorporated into a sequence-to-sequence (S2S) architecture for question generation and its extension with dual decoders to additionally yield poll choices (answers). For experiments, we collect a large-scale Chinese dataset from Sina Weibo containing over 20K polls. The results show that our model outperforms the popular S2S models without exploiting topics from comments and the dual decoder design can further benefit the prediction of both questions and answers. Human evaluations further exhibit our superiority in yielding high-quality polls helpful to draw user engagements.

This paper studies social emotions to online discussion topics. While most prior work focus on emotions from writers, we investigate readers' responses and explore the public feelings to an online topic. A large-scale dataset is collected from Chinese microblog Sina Weibo with over 13 thousand trending topics, emotion votes in 24 fine-grained types from massive participants, and user comments to allow context understanding.(1) In experiments, we examine baseline performance to predict a topic's possible social emotions in a multi-label classification setting. The results show that a seq2seq model with user comment modeling performs the best, even surpassing human prediction. More analyses shed light on the effects of emotion types, topic description lengths, contexts from user comments, and the limited capacity of the existing models.

We present Capacitivo, a contact-based object recognition technique developed for interactive fabrics, using capacitive sensing. Unlike prior work that has focused on metallic objects, our technique recognizes non-metallic objects such as food, different types of fruits, liquids, and other types of objects that are often found around a home or in a workplace. To demonstrate our technique, we created a prototype composed of a 12 x 12 grid of electrodes, made from conductive fabric attached to a textile substrate. We designed the size and separation between the electrodes to maximize the sensing area and sensitivity. We then used a 10-person study to evaluate the performance of our sensing technique using 20 different objects, which yielded a 94.5% accuracy rate. We conclude this work by presenting several different application scenarios to demonstrate unique interactions that are enabled by our technique on fabrics.