Multimodal memory offers a scalable interface for long-video question answering, but existing methods often retrieve captions, frames, transcripts, summaries, or graph facts as isolated fragments. Although searchable, such fragments are not generation-ready: language models must reconstruct cross-modal and temporal alignments at inference time, when context is limited and attribution is difficult. We propose EM²Mem, an event-centric multimodal memory framework that binds heterogeneous evidence to event anchors during memory construction. Each event-indexed memory cell aligns multimodal records, temporal context, graph-linked relations, semantic facts, and provenance, enabling compact evidence readout over grounded multimodal events rather than modality-specific fragments. Across three long-video QA benchmarks, EM²Mem improves average accuracy over the strongest memory baseline by 2.0, 2.4, and 3.7 points, improves strict event-level Top-5 evidence recall by 7.0 points, and reduces per-query latency by 4.67× and total inference tokens by 63.66%.

As LLM agents are deployed in long-horizon sessions, context accumulation drives up inference costs. Existing approaches utilize text pruning or dynamic memory eviction to minimize token footprints; however, their unconstrained sequence mutations alter layouts, introducing prefix mismatches and cache invalidation. This reveals a critical trade-off between text sparsity and prompt cache continuity. To address this, we present **TokenPilot**, a dual-granularity context management framework. Globally, *Ingestion-Aware Compaction* acts as a framework harness to stabilize prompt prefixes and eliminate open-world environmental noise at the ingestion gate. Locally, *Lifecycle-Aware Eviction* monitors the ongoing residual utility of context segments, enforcing a conservative batch-turn schedule to offload content segments only when task relevance expires. Experiments on ‘PinchBench‘ and ‘Claw-Eval‘ under both isolated and continuous modes demonstrate that TokenPilot reduces costs by **61%** and **56%** in isolated mode, and **61%** and **87%** in continuous mode, while maintaining competitive performance compared to prior systems.

Current Large Language Models (LLMs) exhibit a critical modal disconnect: they possess vast semantic knowledge but lack the procedural grounding to respect immutable environmental constraints. Consequently, while these agents implicitly function as world models, their simulations often suffer from physical hallucinations—generating plans that are logically sound but physically unexecutable. Existing alignment strategies predominantly rely on resource-intensive training or fine-tuning, which attempt to compress dynamic environmental rules into static model parameters. However, such parametric encapsulation is inherently rigid, struggling to adapt to the open-ended variability of environmental dynamics without continuous, costly retraining. To solve this fundamental misalignment between semantic planning and environmental reality without relying on parameter updates, we introduce WorldMind, a framework that autonomously constructs a symbolic World Knowledge Repository by synthesizing environmental feedback. Specifically, it unifies Process Experience to enforce semantic physical feasibility via prediction errors and Goal Experience to guide task optimality through successful trajectories. Experiments on EB-ALFRED and EB-Habitat demonstrate that WorldMind achieves the best performance compared to baselines, exhibiting remarkable cross-model and cross-environment transferability.

Long-horizon interactions require language models to manage accumulating information: when to update their state, when to preserve their state, and what to ignore. We study this challenge as Contextual Belief Management (CBM): maintaining a predicted belief state aligned with formal evidence while isolating task-irrelevant noise. To make CBM measurable, we introduce , a closed-world benchmark spanning Rule Discovery and Circuit Diagnosis, where a finite belief space and symbolic verifiers enable exact turn-level evaluation. diagnoses three failures: Failed Stay, Failed Update, and Failed Isolation. Across multiple LLMs, vanilla models exhibit severe CBM failures, while explicit belief-tracking prompts provide limited gains. In contrast, reinforcement learning with belief-state rewards reduces failure rates by 70.9% on average. Further probing reveals latent belief-state dynamics behind these failures, and representation-level steering reduces failure rates by 46.1% across two tasks.

Human beings primarily understand the world through concepts (e.g., dog), abstract mental representations that structure perception, reasoning, and learning. However, how large language models (LLMs) acquire, retain, and forget such concepts during continual pretraining remains poorly understood. In this work, we study how individual concepts are acquired and forgotten, as well as how multiple concepts interact through interference and synergy. We link these behavioral dynamics to LLMs’ internal concept circuits, computational subgraphs associated with specific concepts, and incorporate graph metrics to characterize circuit topology. Our analysis reveals: (1) LLMs’ concept circuits provide a non-trivial, consistent signal of concept learning and forgetting; (2) concept circuit topology evolves stage-wise, rising during concept acquisition, and transiently rising before declining during concept forgetting; (3) semantically similar concepts induce stronger interference than weakly related ones; (4) conceptual knowledge differs in their transferability, with some significantly facilitating the learning of others. Together, our findings provide a circuit-level view of concept learning dynamics and motivate concept-aware training strategies, such as Circuit-aware Experience Replay, which uses circuit topology to prioritize concepts vulnerable to forgetting

Long-term conversational agents need memory systems that capture relationships between events, not merely isolated facts, to support temporal reasoning and multi-hop question answering. Current approaches face a fundamental trade-off: flat memory is efficient but fails to model relational structure, while graph-based memory enables structured reasoning at the cost of expensive and fragile construction. To address these issues, we propose StructMem, a structure-enriched hierarchical memory framework that preserves event-level bindings and induces cross-event connections. By temporally anchoring dual perspectives and performing periodic semantic consolidation, StructMem improves temporal reasoning and multi-hop performance on LoCoMo, while substantially reducing token usage, API calls, and runtime compared to prior memory systems.

As Large Language Models (LLMs) are increasingly deployed in real-world settings, correctness alone is insufficient. Reliable deployment requires maintaining truthful beliefs under contextual perturbations. Existing evaluations largely rely on point-wise confidence like Self-Consistency, which can mask brittle belief. We show that even facts answered with perfect self-consistency can rapidly collapse under mild contextual interference. To address this gap, we propose Neighbor-Consistency Belief (NCB), a structural measure of belief robustness that evaluates response coherence across a conceptual neighborhood. To validate the efficiency of NCB, we introduce a new cognitive stress-testing protocol that probes outputs stability under contextual interference. Experiments across multiple LLMs show that the performance of high-NCB data is relatively more resistant to interference. Finally, we present Structure-Aware Training (SAT), which optimizes context-invariant belief structure and reduces long-tail knowledge brittleness by approximately 30%.

Methods for controlling large language models (LLMs), including local weight fine-tuning, LoRA-based adaptation, and activation-based interventions, are often studied in isolation, obscuring their connections and making comparison difficult. In this work, we present a unified view that frames these interventions as dynamic weight updates induced by a control signal, placing them within a single conceptual framework. Building on this view, we propose a unified preference-utility analysis that separates control effects into preference, defined as the tendency toward a target concept, and utility, defined as coherent and task-valid generation, and measures both on a shared log-odds scale using polarity-paired contrastive examples. Across methods, we observe a consistent trade-off between preference and utility: stronger control increases preference while predictably reducing utility. We further explain this behavior through an activation manifold perspective, in which control shifts representations along target-concept directions to enhance preference, while utility declines primarily when interventions push representations off the model’s valid-generation manifold. Finally, we introduce a new steering approach guided by this analysis that improves preference while better preserving utility.

Model merging has become one of the key technologies for enhancing the capabilities and efficiency of Large Language Models (LLMs). The open-source community has driven model evolution by iteratively merging existing models. However, a principled understanding of the expected gains and underlying factors in model merging remains lacking. In this work, we examine model evolution through continual merging, analogous to biological evolution, and introduce the concept of model kinship, the degree of similarity or relatedness between LLMs. With comprehensive empirical analysis, we find that there is a certain relationship between model kinship and the performance gains after model merging, which can help guide our selection of candidate models. Inspired by this, we propose a new model merging strategy: Top-k Greedy Merging with Model Kinship, which can yield better performance on benchmark datasets. Specifically, we discover that using model kinship as a criterion can assist us in continuously performing model merging, alleviating the degradation (local optima) in model evolution, whereas model kinship can serve as a guide to escape these traps.

Despite exceptional capabilities in knowledge-intensive tasks, Large Language Models (LLMs) face a critical gap in understanding how they internalize new knowledge, particularly how acquired knowledge becomes structurally embedded in their neural computations. We address this issue through the lens of knowledge circuit evolution, identifying computational subgraphs that facilitate knowledge storage and processing. Our systematic analysis of circuit evolution throughout continual pre-training reveals several key findings: (1) the acquisition of new knowledge is influenced by its relevance to pre-existing knowledge; (2) the evolution of knowledge circuits exhibits a distinct phase shift from formation to optimization; (3) the evolution of knowledge circuits follows a deep-to-shallow pattern. These insights not only advance our theoretical understanding of the mechanisms of new knowledge acquisition in LLMs, but also provide potential implications for improving continual pre-training strategies to enhance model performance.

Knowledge Editing (KE) enables the modification of outdated or incorrect information in large language models (LLMs). While existing KE methods can update isolated facts, they often fail to generalize these updates to multi-hop reasoning tasks that rely on the modified knowledge. Through an analysis of reasoning circuits—the neural pathways LLMs use for knowledge-based inference, we find that current layer-localized KE approaches (e.g., MEMIT, WISE), which edit only single or a few model layers, inadequately integrate updated knowledge into these reasoning pathways. To address this limitation, we present CaKE (Circuit-aware Knowledge Editing), a novel method that enhances the effective integration of updated knowledge in LLMs. By only leveraging a few curated data samples guided by our circuit-based analysis, CaKE stimulates the model to develop appropriate reasoning circuits for newly incorporated knowledge. Experiments show that CaKE enables more accurate and consistent use of edited knowledge across related reasoning tasks, achieving an average improvement of 20% in multi-hop reasoning accuracy on the MQuAKE dataset while requiring less memory than existing KE methods.

In this paper, we introduce EasyEdit2, a framework designed to enable plug-and-play adjustability for controlling Large Language Model (LLM) behaviors. EasyEdit2 supports a wide range of test-time interventions, including safety, sentiment, personality, reasoning patterns, factuality, and language features. Unlike its predecessor, EasyEdit2 features a new architecture specifically designed for seamless model steering. It comprises key modules such as the steering vector generator and the steering vector applier, which enable automatic generation and application of steering vectors to influence the model’s behavior without modifying its parameters. One of the main advantages of EasyEdit2 is its ease of use—users do not need extensive technical knowledge. With just a single example, they can effectively guide and adjust the model’s responses, making precise control both accessible and efficient. Empirically, we report model steering performance across different LLMs, demonstrating the effectiveness of these techniques. We have released the source code on https://github.com/zjunlp/EasyEdit along with a demonstration notebook. In addition, we provide an online system at http://easyedit.zjukg.cn/for real-time model steering, and a demo video at https://www.youtube.com/watch?v=AkfoiPfp5rQ.

Chinese, as a linguistic system rich in depth and complexity, is characterized by distinctive elements such as ancient poetry, proverbs, idioms, and other cultural constructs. However, current Large Language Models (LLMs) face limitations in these specialized domains, highlighting the need for the development of comprehensive datasets that can assess, continuously update, and progressively improve these culturally-grounded linguistic competencies through targeted training optimizations. To address this gap, we introduce CKnowEdit, the first-ever Chinese knowledge editing dataset designed to correct linguistic, factual, and logical errors in LLMs. We collect seven types of knowledge from a wide range of sources, including classical texts, idioms, and content from Baidu Tieba Ruozhiba, taking into account the unique polyphony, antithesis, and logical structures inherent in the Chinese language. By analyzing this dataset, we highlight the challenges current LLMs face in mastering Chinese. Furthermore, our evaluation of state-of-the-art knowledge editing techniques reveals opportunities to advance the correction of Chinese knowledge.

Even with their impressive abilities, Large Language Models (LLMs) such as ChatGPT are not immune to issues of factual or logically consistent. Concretely, the key concern is how to seamlessly update those LLMs to correct mistakes without resorting to an exhaustive retraining or continuous training procedure, both of which can demand significant computational resources and time. Thus, the capability to edit LLMs offers an efficient solution to alter a model’s behavior, notably within a distinct area of interest, without negatively impacting its performance on other tasks. Through this tutorial, we strive to acquaint interested NLP researchers with recent and emerging techniques for editing LLMs. Specifically, we aim to present a systematic and current overview of cutting-edge methods, supplemented with practical tools, and unveil new research opportunities for our audiences. All the valuable resources can be accessed at https://github.com/zjunlp/KnowledgeEditingPapers.

Understanding knowledge mechanisms in Large Language Models (LLMs) is crucial for advancing towards trustworthy AGI. This paper reviews knowledge mechanism analysis from a novel taxonomy including knowledge utilization and evolution. Knowledge utilization delves into the mechanism of memorization, comprehension and application, and creation. Knowledge evolution focuses on the dynamic progression of knowledge within individual and group LLMs. Moreover, we discuss what knowledge LLMs have learned, the reasons for the fragility of parametric knowledge, and the potential dark knowledge (hypothesis) that will be challenging to address. We hope this work can help understand knowledge in LLMs and provide insights for future research.

Recently, there has been a growing interest in knowledge editing for Large Language Models (LLMs). Current approaches and evaluations merely explore the instance-level editing, while whether LLMs possess the capability to modify concepts remains unclear. This paper pioneers the investigation of editing conceptual knowledge for LLMs, by constructing a novel benchmark dataset ConceptEdit and establishing a suite of new metrics for evaluation. The experimental results reveal that, although existing editing methods can efficiently modify concept-level definition to some extent, they also have the potential to distort the related instantial knowledge in LLMs, leading to poor performance. We anticipate this work can inspire further progress in understanding LLMs.

This paper investigates using knowledge editing techniques to detoxify Large Language Models (LLMs). We construct a benchmark, SafeEdit, which covers nine unsafe categories with various powerful attack prompts and equips comprehensive metrics for systematic evaluation. We conduct experiments with several knowledge editing approaches, indicating that knowledge editing has the potential to efficiently detoxify LLMs with limited impact on general performance. Then, we propose a simple yet effective baseline, dubbed Detoxifying with Intraoperative Neural Monitoring (DINM), to diminish the toxicity of LLMs within a few tuning steps via only one instance. We further provide an in-depth analysis of the internal mechanism for various detoxifying approaches, demonstrating that previous methods like SFT and DPO may merely suppress the activations of toxic parameters, while DINM mitigates the toxicity of the toxic parameters to a certain extent, making permanent adjustments. We hope that these insights could shed light on future work of developing detoxifying approaches and the underlying knowledge mechanisms of LLMs.

Large Language Models (LLMs) usually suffer from knowledge cutoff or fallacy issues, which means they are unaware of unseen events or generate text with incorrect facts owing to outdated/noisy data. To this end, many knowledge editing approaches for LLMs have emerged – aiming to subtly inject/edit updated knowledge or adjust undesired behavior while minimizing the impact on unrelated inputs. Nevertheless, due to significant differences among various knowledge editing methods and the variations in task setups, there is no standard implementation framework available for the community, which hinders practitioners from applying knowledge editing to applications. To address these issues, we propose EasyEdit, an easy-to-use knowledge editing framework for LLMs. It supports various cutting-edge knowledge editing approaches and can be readily applied to many well-known LLMs such as T5, GPT-J, LlaMA, etc. Empirically, we report the knowledge editing results on LlaMA-2 with EasyEdit, demonstrating that knowledge editing surpasses traditional fine-tuning in terms of reliability and generalization. We have released the source code on GitHub, along with Google Colab tutorials and comprehensive documentation for beginners to get started. Besides, we present an online system for real-time knowledge editing, and a demo video.

Despite the ability to train capable LLMs, the methodology for maintaining their relevancy and rectifying errors remains elusive. To this end, the past few years have witnessed a surge in techniques for editing LLMs, the objective of which is to alter the behavior of LLMs efficiently within a specific domain without negatively impacting performance across other inputs. This paper embarks on a deep exploration of the problems, methods, and opportunities related to model editing for LLMs. In particular, we provide an exhaustive overview of the task definition and challenges associated with model editing, along with an in-depth empirical analysis of the most progressive methods currently at our disposal. We also build a new benchmark dataset to facilitate a more robust evaluation and pinpoint enduring issues intrinsic to existing techniques. Our objective is to provide valuable insights into the effectiveness and feasibility of each editing technique, thereby assisting the community in making informed decisions on the selection of the most appropriate method for a specific task or context.

Previous studies have revealed that vanilla pre-trained language models (PLMs) lack the capacity to handle knowledge-intensive NLP tasks alone; thus, several works have attempted to integrate external knowledge into PLMs. However, despite the promising outcome, we empirically observe that PLMs may have already encoded rich knowledge in their pre-trained parameters but fails to fully utilize them when applying to knowledge-intensive tasks. In this paper, we propose a new paradigm dubbed Knowledge Rumination to help the pre-trained language model utilize that related latent knowledge without retrieving them from the external corpus. By simply adding a prompt like “As far as I know” to the PLMs, we try to review related latent knowledge and inject them back into the model for knowledge consolidation. We apply the proposed knowledge rumination to various language models, including RoBERTa, DeBERTa, and GPT-3. Experimental results on six commonsense reasoning tasks and GLUE benchmarks demonstrate the effectiveness of our proposed approach, which proves that the knowledge stored in PLMs can be better exploited to enhance performance.

Reasoning, as an essential ability for complex problem-solving, can provide back-end support for various real-world applications, such as medical diagnosis, negotiation, etc. This paper provides a comprehensive survey of cutting-edge research on reasoning with language model prompting. We introduce research works with comparisons and summaries and provide systematic resources to help beginners. We also discuss the potential reasons for emerging such reasoning abilities and highlight future research directions. Resources are available at https://github.com/zjunlp/Prompt4ReasoningPapers (updated periodically).

Multimodal named entity recognition and relation extraction (MNER and MRE) is a fundamental and crucial branch in information extraction. However, existing approaches for MNER and MRE usually suffer from error sensitivity when irrelevant object images incorporated in texts. To deal with these issues, we propose a novel Hierarchical Visual Prefix fusion NeTwork (HVPNeT) for visual-enhanced entity and relation extraction, aiming to achieve more effective and robust performance. Specifically, we regard visual representation as pluggable visual prefix to guide the textual representation for error insensitive forecasting decision. We further propose a dynamic gated aggregation strategy to achieve hierarchical multi-scaled visual features as visual prefix for fusion. Extensive experiments on three benchmark datasets demonstrate the effectiveness of our method, and achieve state-of-the-art performance.