论文检索

Applications · Language, Speech and Dialog

Lihu Chen, Gerard de Melo, Fabian Suchanek, Gael Varoquaux

It is important for Large Language Models (LLMs) to be aware of the boundary of their knowledge, i.e., the mechanism of identifying known and unknown queries. This type of awareness enables models to perform adaptive inference, such as invoking retrieval-augmented generation (RAG), engaging in slow and deep thinking, or abstaining from answering when appropriate. These mechanisms are beneficial to the development of efficient and trustworthy AI. In this work, we propose a method to detect knowledge boundaries via \textbf{\emph{Query-Level Uncertainty }}, which estimates if a model is capable of to answering a given query before generating any tokens. To this end, we propose a novel, training-free method called \textbf{\emph{Internal Confidence}}, which leverages self-evaluations across layers and tokens to provide a reliable signal of uncertainty. Empirical studies on both factual question answering and mathematical reasoning tasks demonstrate that our internal confidence can outperform several baselines. Furthermore, we showcase that our proposed method can be used for adaptive inference, such as efficient RAG and model cascading, thereby reducing inference costs while preserving overall performance.

Applications · Language, Speech and Dialog

Minhua Lin, Enyan Dai, Hui Liu, Xianfeng Tang, Yuliang Yan, Zhenwei Dai, Jingying Zeng, Zhiwei Zhang, Fali Wang, Hongcheng Gao 等

As Large Language Models (LLMs) are increasingly applied in high-stakes domains, their ability to reason strategically under uncertainty becomes critical. Poker provides a rigorous testbed, requiring not only strong actions but also principled, game-theoretic reasoning. In this paper, we conduct a systematic study of LLMs in multiple realistic poker tasks, evaluating both gameplay outcomes and reasoning traces. Our analysis reveals LLMs fail to compete against traditional algorithms and identifies three recurring flaws: reliance on heuristics, factual misunderstandings, and a “knowing–doing” gap where actions diverge from reasoning. An initial attempt with behavior cloning and step-level reinforcement learning improves reasoning style but remains insufficient for accurate game-theoretic play. Motivated by these limitations, we propose ToolPoker, a tool-integrated reasoning framework that combines external solvers for GTO-consistent actions with more precise professional-style explanations. Experiments demonstrate that ToolPoker achieves state-of-the-art gameplay while producing reasoning traces that closely reflect game-theoretic principles.

Deep Learning · Everything Else

Rubin Wei, Jiaqi Cao, Jiarui Wang, Jushi Kai, Qipeng Guo, Bowen Zhou, Zhouhan Lin

Modern approaches to enhancing Large Language Models' factual accuracy and knowledge utilization face a fundamental trade-off: non-parametric retrieval-augmented generation (RAG) provides flexible access to external knowledge but suffers from high inference latency and shallow integration, while parametric fine-tuning methods like LoRA risk catastrophic forgetting and degraded general capabilities. In this work, we propose MLP Memory, a lightweight parametric module that learns to internalize retrieval patterns without explicit document access. By pretraining an MLP to imitate a $k$NN retriever's behavior on the entire pretraining dataset, we create a differentiable memory component that captures the benefits of retrieval-based knowledge access in a fully parametric form. Our architecture integrates this pretrained MLP Memory with Transformer decoders through simple probability interpolation, achieving 12.3\% relative improvement on five question-answering benchmarks and 5.2 points absolute gain across nine general NLP tasks, while reducing hallucinations by up to 10 points on HaluEval. Moreover, MLP Memory delivers 2.5$\times$ faster inference than RAG with superior accuracy. Our findings show that learning retrieval patterns parametrically bridges the gap between efficient inference and effective knowledge access, offering a practical alternative to both RAG and fine-tuning approaches.

Social Aspects · Trustworthy Machine Learning

Wenbo Pan, Jie Xu, Qiguang Chen, Junhao Dong, Libo Qin, Xinfeng Li, Yu Haining, Xiaohua Jia

Large Language Models (LLMs) should refuse to answer questions beyond their knowledge. This capability, which we term knowledge-aware refusal, is crucial for factual reliability, while existing metrics fail to capture this ability. In this work, we propose the Refusal Index (RI), a novel and principled metric that measures how accurately LLMs refuse questions they do not know. We define RI as Spearman's rank correlation between refusal probability and error probability. RI is practically measurable with a lightweight two-pass evaluation method which only require observed refusal rates across two standard evaluation runs. Extensive experiments across 16 models and 5 datasets demonstrate that RI accurately quantifies a model's knowledge-aware refusal capability. Notably, RI remains stable across different refusal rates and provides consistent model rankings independent of a model's overall accuracy and refusal rates. These properties suggest RI captures a stable, intrinsic aspect of model knowledge calibration. More importantly, RI provides insight into an important but previously overlooked aspect of LLM factuality: while LLMs achieve high accuracy on factual tasks, their refusal behavior can be unreliable and fragile.

Stella Li, Avinandan Bose, Faeze Brahman, Simon Du, Pang Wei Koh, Maryam Fazel, Yulia Tsvetkov

Current large language model (LLM) development treats task-solving and preference-alignment as separate challenges, optimizing first for objective correctness, then for alignment to aggregated human preferences. This paradigm fails in human-facing applications where solving a problem correctly is insufficient if the response mismatches the user’s needs. This challenge intensifies in just-in-time scenarios where no prior user interaction history exists due to cold-start conditions or privacy constraints. LLMs need to proactively identify what they don’t know about the user, strategically elicit preference values through questioning, then adapt their reasoning processes and responses accordingly—a complicated chain of cognitive processes which we term personalized reasoning. We introduce PREFDISCO, an evaluation methodology that transforms static benchmarks into interactive personalization tasks using psychologically-grounded personas with sparse, context-dependent preferences, and define PREFALIGN as a fine-grained rubric-based metric for measuring preference alignment. PREFDISCO builds scenarios where identical questions require different reasoning chains depending on user context, as optimal explanation approaches vary by individual expertise and preferences while maintaining factual accuracy. Evaluation of 21 frontier models across 10 tasks reveals 29.0% of naive personalization attempts produce worse preference alignment than generic responses, yet generic responses also fail to serve individual user needs. These findings suggest personalized reasoning requires dedicated development rather than emerging naturally. PREFDISCO provides a foundation for developing systems that can adapt to individual users in education, healthcare, and technical domains where personalization is critical.

Social Aspects · Trustworthy Machine Learning

Yongxin Deng, Zhen Fang, Sharon Li, Ling Chen

Hallucination detection is critical for deploying large language models (LLMs) in real-world applications. Existing hallucination detection methods achieve strong performance when the training and test data come from the same domain, but they suffer from poor cross-domain generalization. In this paper, we study an important yet overlooked problem, termed generalizable hallucination detection (GHD), which aims to train hallucination detectors on data from a single domain while ensuring robust performance across diverse related domains. In studying GHD, we simulate multi-turn dialogues following LLMs' initial response and observe an interesting phenomenon: hallucination-initiated multi-turn dialogues universally exhibit larger uncertainty fluctuations than factual ones across different domains. Based on the phenomenon, we propose a new score SpikeScore, which quantifies abrupt fluctuations in multi-turn dialogues. Through both theoretical analysis and empirical validation, we demonstrate that SpikeScore achieves strong cross-domain separability between hallucinated and non-hallucinated responses. Experiments across multiple LLMs and benchmarks demonstrate that the SpikeScore-based detection method outperforms representative baselines in cross-domain generalization and surpasses advanced generalization-oriented methods, verifying the effectiveness of our method in cross-domain hallucination detection.

Deep Learning · Generative Models and Autoencoders

Todd Nief, David Reber, Sean Richardson, Ari Holtzman

When an LLM learns a new fact during finetuning (e.g., new movie releases, newly elected pope, etc.), where does this information go? Are entities enriched with relation information immediately, or do models recall information just-in-time before a prediction? Or, are "all of the above" true, with LLMs implementing multiple redundant heuristics? Existing localization approaches (e.g., activation patching) are ill-suited for this analysis because they usually replace parts of the residual stream, thus overriding previous information. To fill this interpretability gap, we propose dynamic weight grafting, an analysis technique that selectively grafts subsets of weights from a finetuned model onto a pretrained model. Using this technique, we show two separate pathways for retrieving finetuned relation information: 1) "enriching" the residual stream with relation information while processing the tokens that correspond to an entity (e.g., "Zendaya" in "Zendaya co-starred with Timothée Chalamet" and 2) "recalling" this information at the final token position before generating a target fact. In some cases, models need information from both of these pathways to correctly generate finetuned facts while, in other cases, either the "enrichment" or "recall" pathway alone is sufficient. We localize the "recall" pathway to model components---finding that "recall" occurs via both task-specific attention mechanisms and an entity-specific extraction step in the feedforward networks of the final layers before prediction. By targeting model components and parameters, as opposed to just activations, we are able to understand the mechanisms by which finetuned knowledge is retrieved during generation.

Theory · Interpretability and Visualization

Jiayu Yang, Yuxuan Fan, Songning Lai, Shengen Wu, Jiaqi Tang, Chun Kang, Zhijiang Guo, Yutao Yue

LLMs require efficient knowledge editing (KE) to update factual information, yet existing methods exhibit significant performance decay in multi-hop factual recall. This failure is particularly acute when edits involve intermediate implicit subjects within reasoning chains. Through causal analysis, we reveal that this limitation stems from an oversight of how chained knowledge is dynamically represented and utilized at the neuron level. We discover that during multi-hop reasoning, implicit subjects function as query neurons, which sequentially activate corresponding value neurons across transformer layers to accumulate information toward the final answer—a dynamic prior KE work has overlooked. Guided by this insight, we propose ACE (Attribution-Controlled Knowledge Editing), a framework that leverages neuron-level attribution to identify and edit these critical query-value (Q-V) pathways. Ace provides a mechanistically grounded solution for multi-hop KE, empirically outperforming state-of-the-art methods by 9.44% on GPT-J and 37.46% on Qwen3-8B. Our analysis further reveals more fine-grained activation patterns in Qwen3 and demonstrates that the semantic interpretability of value neurons is orchestrated by query-driven accumulation. These findings establish a new pathway for advancing KE capabilities based on the principled understanding of internal reasoning mechanisms.

Social Aspects · Accountability, Transparency and Interpretability

Yupu Gu, Rongzhe Wei, Andy Zhu, Pan Li

Knowledge editing (KE) is crucial for making precise modifications to factual knowledge within large language models (LLMs). Existing KE methods, however, are primarily designed for dense architectures, limiting their applicability to the increasingly popular sparse Mixture-of-Experts (MoE) models that power modern scalable LLMs. While MoEs offer remarkable efficiency and capacity scaling, their unique structure introduces new challenges for KE. Naively adapting dense-model editors to MoEs is not only computationally expensive but also induces routing distribution shifts that degrade model stability and consistency. To address these challenges, we introduce MoEEdit, the first systematic framework for routing-stable knowledge editing in MoE LLMs. Our approach reparameterizes expert updates through per-expert null-space projections, ensuring router inputs remain invariant to suppress these shifts, and solves the resulting block-structured optimization with an efficient block coordinate descent (BCD) solver. Experiments demonstrate that MoEEdit achieves state-of-the-art efficacy and generalization, while maintaining high specificity, routing stability, and superior computational and memory efficiency. Our work establishes a robust foundation for scalable and precise knowledge editing in modern sparse LLMs by highlighting the necessity of routing-stable interventions.

Deep Learning · Generative Models and Autoencoders

Qinyuan Wu, Soumi Das, Mahsa Amani, Bishwamittra Ghosh, Mohammad Aflah Khan, Krishna Gummadi, Muhammad Bilal Zafar

Rote learning is a memorization technique based on repetition. Many researchers argue that rote learning hinders generalization because it encourages verbatim memorization rather than deeper understanding. This concern extends even to factual knowledge, which inevitably requires a certain degree of memorization. In this work, we challenge this view and demonstrate that large language models (LLMs) can, in fact, generalize over rote memorized data. We introduce a two-phase “memorize-then-generalize” framework, where the model first rote memorizes factual subject-object associations using a synthetic semantically meaningless key token and then learns to generalize by fine-tuning on a small set of semantically meaningful prompts. Extensive experiments over 8 LLMs show that the models can reinterpret rote memorized data through the semantically meaningful prompts, as evidenced by the emergence of structured, semantically aligned latent representations between the key token and the semantically meaningful prompts. This surprising finding opens the door to both effective and efficient knowledge injection as well as possible risks of repurposing the memorized data for malicious usage.

Reinforcement Learning · Deep RL

Jiajin Tang, Yang Gao, Wenjie Wang, Sibei Yang, Xing Chen

With the rapid advancement of data science, charts have evolved from simple numerical presentation tools to essential instruments for insight discovery and decision-making support. However, current chart data intelligence exhibits significant limitations in deep research capabilities, with existing methods predominantly addressing shallow tasks such as visual recognition or factual question-answering, rather than the complex reasoning and high-level data analysis that deep research requires. This limitation stems from two primary technical bottlenecks: at the training level, existing post-training techniques exhibit deficiencies in handling multi-dimensional reward signal interference and heterogeneous data gradient conflicts, preventing models from achieving balanced development across multiple capability dimensions; at the evaluation level, current methods remain limited to factual retrieval and basic computation, failing to assess end-to-end analytic reasoning and other deep research capabilities. To address the training challenge, we propose PRPO, which performs parallel optimization across reward dimensions and capability partitioning across data types, effectively disentangling conflicts between heterogeneous data and multi-dimensional reward signals while ensuring optimization stability. For the evaluation challenge, we construct MCDR-Bench based on the ``error uniqueness principle," transforming subjective generation assessment into objective error identification through controllable error injection, enabling quantifiable evaluation of deep research capabilities. Experimental validation confirms that the proposed PRPO and MCDR-Bench jointly establish a unified framework that systematically advances chart deep research through enhanced collaborative training and objective evaluation.

Social Aspects · Trustworthy Machine Learning

Robert Adrian Minut, Hazem Dewidar, Iacopo Masi

We reinterpret the final Large Language Model (LLM) softmax classifier as an Energy-Based Model (EBM), decomposing the sequence-to-sequence probability chain into multiple interacting EBMs at inference. This principled approach allows us to track “energy spills” during decoding, which we empirically show correlate with factual errors, biases, and failures. Similar to Orgad et al. (2025), our method localizes the exact answer token and subsequently tests for hallucinations. Crucially, however, we achieve this without requiring trained probe classifiers or activation ablations. Instead, we introduce two completely training-free metrics derived directly from output logits: **spilled energy**, which captures the discrepancy between energy values across consecutive generation steps that should theoretically match, and **marginalized energy**, which is measurable at a single step. Evaluated on nine benchmarks across state-of-the-art LLMs (including LLaMA, Mistral, and Gemma) and on synthetic algebraic operations (Qwen3), our approach demonstrates robust, competitive hallucination detection and cross-task generalization. Notably, these results hold for both pretrained and instruction-tuned variants without introducing any training overhead. Code available at [github.com/OmnAI-Lab/spilled-energy/](https://github.com/OmnAI-Lab/spilled-energy/)

Applications · Everything Else

Xuechen Zhang, Koustava Goswami, Samet Oymak, Jiasi Chen, Nedim Lipka

Retrieval-augmented generation (RAG) has strong potential for producing accurate and factual outputs by combining language models (LMs) with evidence retrieved from large text corpora. However, current pipelines are limited by static chunking and flat retrieval: documents are split into short, predetermined, fixed-size chunks, embeddings are retrieved uniformly, and generation relies on whatever chunks are returned. This design brings challenges, as retrieval quality is highly sensitive to chunk size, often introduces noise from irrelevant or misleading chunks, and scales poorly to large corpora. We present SmartChunk retrieval, a query-adaptive framework for efficient and robust long-document question answering (QA). SmartChunk uses (i) a planner that predicts the optimal chunk abstraction level for each query, and (ii) a lightweight compression module that produces high-level chunk embeddings without repeated summarization. By adapting retrieval granularity on the fly, SmartChunk balances accuracy with efficiency and avoids the drawbacks of fixed strategies. Notably, our planner can reason about chunk abstractions through a novel reinforcement learning scheme, STITCH, which boosts accuracy and generalization. To reflect real-world applications, where users face diverse document types and query styles, we evaluate SmartChunk on five QA benchmarks plus one out-of-domain dataset. Across these evaluations, SmartChunk outperforms state-of-the-art RAG baselines, while reducing cost. Further analysis demonstrates strong scalability with larger corpora and consistent gains on out-of-domain datasets, highlighting its effectiveness as a general framework for adaptive retrieval.

Applications · Language, Speech and Dialog

Yanchen Jiang, Zhe Feng, Aranyak Mehta

Large language models (LLMs) are increasingly used in modern search and answer systems to synthesize multiple, sometimes conflicting, texts into a single response, yet current pipelines offer weak incentives for sources to be accurate and are vulnerable to adversarial content. We introduce Truthful Text Summarization (TTS), an incentive-aligned framework that improves factual robustness without ground-truth labels. TTS (i) decomposes a draft synthesis into atomic claims, (ii) elicits each source’s stance on every claim, (iii) scores sources with an adapted multi-task peer-prediction mechanism that rewards informative agreement, and (iv) filters unreliable sources before re-summarizing. We establish formal guarantees that align a source’s incentives with informative honesty, making truthful reporting the utility-maximizing strategy. Experiments show that TTS improves factual accuracy and robustness while preserving fluency, aligning exposure with informative corroboration and disincentivizing manipulation.

Deep Learning · Generative Models and Autoencoders

Christopher Ackerman

The possibility of LLM self-awareness and even sentience is gaining increasing public attention and has major safety and policy implications, but the science of measuring them is still in a nascent state. Here we introduce a novel methodology for quantitatively evaluating metacognitive abilities in LLMs. Taking inspiration from research on metacognition in nonhuman animals, our approach eschews model self-reports and instead tests to what degree models can strategically deploy knowledge of internal states. Using two experimental paradigms, we demonstrate that frontier LLMs introduced since early 2024 show increasingly strong evidence of certain metacognitive abilities, specifically the ability to assess and utilize their own confidence in their ability to answer factual and reasoning questions correctly and the ability to anticipate what answers they would give and utilize that information appropriately. We buttress these behavioral findings with an analysis of the token probabilities returned by the models, which suggests the presence of an upstream internal signal that could provide the basis for metacognition. We further find that these abilities 1) are limited in resolution, 2) emerge in context-dependent manners, and 3) seem to be qualitatively different from those of humans. We also report intriguing differences across models of similar capabilities, suggesting that LLM post-training may have a role in developing metacognitive abilities.

Deep Learning · Everything Else

Haewon Park, Sangwoo Kim, Yohan Jo

Knowledge editing aims to efficiently correct factual errors in language models. Widely used locate-then-edit methods update an MLP layer by adjusting its weights to change the mapping between the layer’s input vector (key) and output vector (value), thereby editing the model’s knowledge. As this update is driven by key and value vectors, obtaining these vectors without careful constraints causes significant model perturbations beyond the targeted edit, a common issue in many prior knowledge editing methods. To address this, we propose Subspace Knowledge Edit (SUIT), which computes key and value vectors only within the subspace of critical features relevant to the edit. Our empirical results on LLaMA3, GPT-J, and Qwen2.5 models show that SUIT dramatically improves knowledge preservation over strong baselines while maintaining high editing performance. These results support the claim that SUIT successfully identifies the critical subspace for the edit. Beyond quantitative gains, our analyses show that SUIT reduces unintended perturbations in hidden states while confining updates to directions that are more effective for editing. Taken together, these findings establish edit-critical subspace identification as a key principle for reliable, low-perturbation knowledge editing. Our code is available at https://github.com/holi-lab/SUIT.

Computer Vision · Vision Models & Multimodal

Kai Mei, Jiang Guo, Shuaichen Chang, Mingwen Dong, Dongkyu Lee, Xing Niu, Jiarong Jiang

Large Language Models (LLMs) can serve as world models to enhance agent decision-making in digital environments by simulating future states and predicting action outcomes, potentially eliminating costly trial-and-error exploration. However, this capability is fundamentally limited by LLM's tendency to hallucination and their reliance on static training knowledge, which could lead to compounding errors that inhibit long-horizon simulations. To systematically investigate whether LLMs are appropriate for world modeling, we probe two core capabilities of world models -- *future state prediction* and *reward estimation* -- through three tasks: next-state identification, full-procedure planning alignment, and milestone transition recognition. Our analysis shows that while LLMs effectively capture immediate next states and identify meaningful state transitions, their performance rapidly degrades in full-procedure planning. This highlights LLMs’ limitations in reliably modeling environment dynamics over long horizons. To address these limitations, we propose the Retrieval-augmented World Model (R-WoM), which grounds LLM simulations by incorporating factual, up-to-date knowledge retrieved from external tutorials. Experiments show that R-WoM achieves relative improvements of up to 23.4\% and 16.3\% on the subsets of OSWorld and Webarena compared to baselines, with particular advantage in longer-horizon simulations.

Computer Vision · Image and Video Generation

Le Zhuo, Songhao Han, Yuandong Pu, Boxiang Qiu, Sayak Paul, Yue Liao, Yihao Liu, Jie Shao, Xi Chen, Si Liu 等

While modern visual generation models excel at creating aesthetically pleasing natural images, they struggle with producing or editing structured visuals like charts, diagrams, and mathematical figures, which demand composition planning, text rendering, and multimodal reasoning for factual fidelity. To address this, we present the first comprehensive, systematic investigation of this domain, encompassing data construction, model training, and an evaluation benchmark. First, we construct a large-scale dataset of 1.3 million high-quality structured image pairs derived from executable drawing programs and augmented with chain-of-thought reasoning annotations. Building on it, we train a unified model that integrates a VLM with FLUX.1 Kontext via a lightweight connector for enhanced multimodal understanding. A three-stage training curriculum enables progressive feature alignment, knowledge infusion, and reasoning-augmented generation, further boosted by an external reasoner at inference time. Finally, we introduce StructBench, a novel benchmark for generation and editing with over 1,700 challenging instances, and an accompanying evaluation metric, StructScore, which employs a multi-round Q\&A protocol to assess fine-grained factual accuracy. Evaluations of 15 models reveal that even leading closed-source systems remain far from satisfactory. Our model attains strong editing performance, and inference-time reasoning yields consistent gains across diverse architectures. By releasing the dataset, model, and benchmark, we aim to advance unified multimodal foundations for structured visuals.

Social Aspects · Fairness, Equity, Justice and Safety

Jiaying Wu, Fanxiao Li, Zihang Fu, Min-Yen Kan, Bryan Hooi

The impact of multimodal misinformation arises not only from factual inaccuracies but also from the misleading narratives that creators deliberately embed. Interpreting such creator intent is therefore essential for multimodal misinformation detection (MMD) and effective information governance. To this end, we introduce DeceptionDecoded, a large-scale benchmark of 12,000 image-caption pairs grounded in trustworthy reference articles, created using an intent-guided simulation framework that models both the desired influence and the execution plan of news creators. The dataset captures both misleading and non-misleading cases, spanning manipulations across visual and textual modalities, and supports three intent-centric tasks: (1) misleading intent detection, (2) misleading source attribution, and (3) creator desire inference. We evaluate 14 state-of-the-art vision-language models (VLMs) and find that they struggle with intent reasoning, often relying on shallow cues such as surface-level alignment, stylistic polish, or heuristic authenticity signals. To bridge this, our framework systematically synthesizes data that enables models to learn implication-level intent reasoning. Models trained on DeceptionDecoded demonstrate strong transferability to real-world MMD, validating our framework as both a benchmark to diagnose VLM fragility and a data synthesis engine that provides high-quality, intent-focused resources for enhancing robustness in real-world multimodal misinformation governance.

General Machine Learning · Everything Else

Soyeon Kim, Jindong Wang, Xing Xie, Steven Whang

Facts change over time, making it essential for Large Language Models (LLMs) to handle time-sensitive factual knowledge accurately and reliably. Although factual Time-Sensitive Question-Answering (TSQA) tasks have been widely developed, existing benchmarks often face manual bottlenecks that limit scalable and comprehensive TSQA evaluation. To address this issue, we propose TDBench, a new benchmark that systematically constructs TSQA pairs by harnessing temporal databases and database techniques, such as temporal functional dependencies, temporal SQL, and temporal joins. We also introduce a new evaluation metric called time accuracy, which assesses the validity of time references in model explanations alongside traditional answer accuracy for a more fine-grained TSQA evaluation. Extensive experiments on contemporary LLMs show how TDBench enables scalable and comprehensive TSQA evaluation while reducing the reliance on human labor, complementing current TSQA evaluation approaches that largely center on Wikipedia/Wikidata by enabling LLM evaluation on application-specific data.