Daily Papers

Coding agents are rapidly changing the landscape of software development, moving from inline completion to autonomous systems that edit repositories, open pull requests, respond to issues, and run scheduled or webhook triggered routines across the development life cycle. The next generation is increasingly described as proactive and long-horizon: agents should notice relevant changes before the developer asks, connect signals across tools, decide when to interrupt, and carry preferences across sessions. Yet the field still lacks a clear account of what proactivity means for software development, how it differs from autonomy, what acceptance criteria proactive long-horizon tasks should satisfy, and which metrics determine whether unsolicited agent behavior is useful rather than merely active. Proactive coding agents should be evaluated by the quality and improvement of their insight policy: the policy that decides what matters next, what evidence supports it, whether to show it, and how to adapt after feedback. This view is grounded in the principles of mixed initiative interaction. We propose a three level taxonomy of proactivity (Reactive, Scheduled, and Situation Aware), compare contemporary coding agents against five practical criteria, and sketch an active user simulation protocol with three evaluation targets: Insight Decision Quality (IDQ), Context Grounding Score (CGS), and Learning Lift

Large Language Models (LLMs) have demonstrated considerable success in open-book question answering (QA), where the task requires generating answers grounded in a provided external context. A critical challenge in open-book QA is to ensure that model responses are based on the provided context rather than its parametric knowledge, which can be outdated, incomplete, or incorrect. Existing evaluation methods, primarily based on the LLM-as-a-judge approach, face significant limitations, including biases, scalability issues, and dependence on costly external systems. To address these challenges, we propose a novel metric that contrasts the perplexity of the model response under two conditions: when the context is provided and when it is not. The resulting score quantifies the extent to which the model's answer relies on the provided context. The validity of this metric is demonstrated through a series of experiments that show its effectiveness in identifying whether a given answer is grounded in the provided context. Unlike existing approaches, this metric is computationally efficient, interpretable, and adaptable to various use cases, offering a scalable and practical solution to assess context utilization in open-book QA systems.

Spec-driven development (SDD) with AI coding agents provides a structured workflow, but agents often remain "context blind" in large, evolving repositories, leading to hallucinated APIs and architectural violations. We present Spec Kit Agents, a multi-agent SDD pipeline (with PM and developer roles) that adds phase-level, context-grounding hooks. Read-only probing hooks ground each stage (Specify, Plan, Tasks, Implement) in repository evidence, while validation hooks check intermediate artifacts against the environment. We evaluate 128 runs covering 32 features across five repositories. Context-grounding hooks improve judged quality by +0.15 on a 1-5 composite LLM-as-judge score (+3.0 percent of the full score; Wilcoxon signed-rank, p < 0.05) while maintaining 99.7-100 percent repository-level test compatibility. We further evaluate the framework on SWE-bench Lite, where augmentation hooks improve baseline by 1.7 percent, achieving 58.2 percent Pass@1.

We propose a new long-context financial benchmark, FailSafeQA, designed to test the robustness and context-awareness of LLMs against six variations in human-interface interactions in LLM-based query-answer systems within finance. We concentrate on two case studies: Query Failure and Context Failure. In the Query Failure scenario, we perturb the original query to vary in domain expertise, completeness, and linguistic accuracy. In the Context Failure case, we simulate the uploads of degraded, irrelevant, and empty documents. We employ the LLM-as-a-Judge methodology with Qwen2.5-72B-Instruct and use fine-grained rating criteria to define and calculate Robustness, Context Grounding, and Compliance scores for 24 off-the-shelf models. The results suggest that although some models excel at mitigating input perturbations, they must balance robust answering with the ability to refrain from hallucinating. Notably, Palmyra-Fin-128k-Instruct, recognized as the most compliant model, maintained strong baseline performance but encountered challenges in sustaining robust predictions in 17% of test cases. On the other hand, the most robust model, OpenAI o3-mini, fabricated information in 41% of tested cases. The results demonstrate that even high-performing models have significant room for improvement and highlight the role of FailSafeQA as a tool for developing LLMs optimized for dependability in financial applications. The dataset is available at: https://huggingface.co/datasets/Writer/FailSafeQA

Real-world tasks often require models to learn from task-specific context rather than relying only on pre-trained knowledge. While recent work has highlighted this capability as context learning, existing evaluations mainly focus on textual contexts. In many practical settings, however, the context to be learned from is multimodal: scientific findings are conveyed through figures and tables, financial indicators are scattered across converted reports, and spatial decisions depend on maps, scenes, or web pages. We introduce CLBench-V, a benchmark for multimodal context learning that addresses the difficulty of localizing where context use breaks down by organizing tasks around three dimensions: context grounding, new information application, and new knowledge learning. CLBench-V combines converted public benchmarks with newly constructed datasets spanning domains such as science, finance, long-document understanding, spatial reasoning, and web-based visual question answering. To reduce the cost of constructing domain-specific context-learning tasks, we further use automated construction and filtering procedures for our newly built datasets. Across 3,443 instances and six recent multimodal models, the best overall score is only 0.2847, indicating that multimodal context learning remains far from saturated. Moreover, InternVL3.5-30B-A3B performs best on context grounding and new knowledge learning, while Qwen3.5-Plus performs best on new information application. We further analyze judge reliability, context length, image count, and representative failure cases. Code is available at https://github.com/IamLihua/CLBench-V.

When retrieval-augmented generation (RAG) systems hallucinate, what geometric trace does this leave in embedding space? We introduce the Semantic Grounding Index (SGI), defined as the ratio of angular distances from the response to the question versus the context on the unit hypersphere S^{d-1}.Our central finding is semantic laziness: hallucinated responses remain angularly proximate to questions rather than departing toward retrieved contexts. On HaluEval (n=5,000), we observe large effect sizes (Cohen's d ranging from 0.92 to 1.28) across five embedding models with mean cross-model correlation r=0.85. Crucially, we derive from the spherical triangle inequality that SGI's discriminative power should increase with question-context angular separation θ(q,c)-a theoretical prediction confirmed empirically: effect size rises monotonically from d=0.61 -low θ(q,c), to d=1.27 -high θ(q,c), with AUC improving from 0.72 to 0.83. Subgroup analysis reveals that SGI excels on long responses (d=2.05) and short questions (d=1.22), while remaining robust across context lengths. Calibration analysis yields ECE=0.10, indicating SGI scores can serve as probability estimates, not merely rankings. A critical negative result on TruthfulQA (AUC=0.478) establishes that angular geometry measures topical engagement rather than factual accuracy. SGI provides computationally efficient, theoretically grounded infrastructure for identifying responses that warrant verification in production RAG deployments.

We propose Context-aware Video-text Alignment (CVA), a novel framework to address a significant challenge in video temporal grounding: achieving temporally sensitive video-text alignment that remains robust to irrelevant background context. Our framework is built on three key components. First, we propose Query-aware Context Diversification (QCD), a new data augmentation strategy that ensures only semantically unrelated content is mixed in. It builds a video-text similarity-based pool of replacement clips to simulate diverse contexts while preventing the ``false negative" caused by query-agnostic mixing. Second, we introduce the Context-invariant Boundary Discrimination (CBD) loss, a contrastive loss that enforces semantic consistency at challenging temporal boundaries, making their representations robust to contextual shifts and hard negatives. Third, we introduce the Context-enhanced Transformer Encoder (CTE), a hierarchical architecture that combines windowed self-attention and bidirectional cross-attention with learnable queries to capture multi-scale temporal context. Through the synergy of these data-centric and architectural enhancements, CVA achieves state-of-the-art performance on major VTG benchmarks, including QVHighlights and Charades-STA. Notably, our method achieves a significant improvement of approximately 5 points in Recall@1 (R1) scores over state-of-the-art methods, highlighting its effectiveness in mitigating false negatives.

Retrieval-augmented generation (RAG) reduces but does not eliminate hallucination, and existing detectors return a single answer-level score that does not indicate which sentence is unsupported, or why. To close this gap, we introduce Grounding-Aware Sensitivity by Perturbation (GASP), a span-level detector that scores each answer sentence by how strongly its likelihood depends on the retrieved evidence, a quantity we term grounding sensitivity. GASP holds the answer fixed and re-scores it under the full context, under no context, and with each chunk removed, then measures the log-likelihood drops and Jensen-Shannon divergences (JSD). The likelihood of a grounded sentence collapses once its supporting passage is removed, whereas a hallucinated sentence is almost unaffected, a contrast we interpret by casting decoding as a random nonlinear iterated function system (RNIFS). We evaluate GASP on three benchmarks (RAGTruth, TofuEval, RAGBench) with three instruction-tuned scorers from two model families (Qwen2.5-0.5B, Qwen2.5-1.5B, and SmolLM2-1.7B) under a leakage-clean protocol. On RAGTruth it reaches a response-level area under the ROC curve (AUC) of about 0.73 and a span-level AUC of about 0.67, improving significantly over perplexity and by clear margins over length, whole-context natural language inference (NLI), and self-consistency baselines. The only baseline competitive at the span level is a well-configured chunk-level entailment verifier, which requires a separate model, whereas a training-free threshold on the grounding features matches the trained classifier without labeled data and serves as the default detector. Beyond RAGTruth, the signal transfers to TofuEval but not to short-answer question answering in RAGBench, showing GASP is best suited to outputs constructed from the retrieved context rather than answers recoverable from parametric knowledge.

Evaluating whether multimodal large language models truly understand long-form scientific papers remains challenging: answer-only metrics and synthetic "Needle-In-A-Haystack" tests often reward answer matching without requiring a causal, evidence-linked reasoning trace in the document. We propose the "Fish-in-the-Ocean" (FITO) paradigm, which requires models to construct explicit cross-modal evidence chains within native scientific documents. To operationalize FITO, we build SIN-Data, a scientific interleaved corpus that preserves the native interleaving of text and figures. On top of it, we construct SIN-Bench with four progressive tasks covering evidence discovery (SIN-Find), hypothesis verification (SIN-Verify), grounded QA (SIN-QA), and evidence-anchored synthesis (SIN-Summary). We further introduce "No Evidence, No Score", scoring predictions when grounded to verifiable anchors and diagnosing evidence quality via matching, relevance, and logic. Experiments on eight MLLMs show that grounding is the primary bottleneck: Gemini-3-pro achieves the best average overall score (0.573), while GPT-5 attains the highest SIN-QA answer accuracy (0.767) but underperforms on evidence-aligned overall scores, exposing a gap between correctness and traceable support.

Multi-video event understanding demands models that can locate and attribute query-relevant evidence scattered across long, heterogeneous video corpora. Existing large vision-language models (LVLMs) often underperform in this regime because they quickly exhaust their context budget and struggle to precisely localize evidentially important segments, frequently missing dense informational cues such as broadcast graphics, subtitles, and scoreboards. We introduce TRACE, an evidence grounding-guided framework that follows a ground-before-reasoning strategy for multi-video event reasoning. Our approach first builds a structured, text-searchable timeline for each video using OCR and object detection. A text-only LLM then conducts query-aware evidence localization, selecting relevant moments prior to any downstream visual reasoning. The retrieved frames and their grounding summaries are subsequently used to steer LVLM-based claim generation and cross-video citation consolidation. Experiments on MAGMaR 2026 and WikiVideo demonstrate that structured grounding markedly boosts factual completeness and attribution fidelity. On the MAGMaR validation split, TRACE raises macro-average MiRAGE F1 from 0.705 to 0.811 compared to an unguided Qwen3-VL-30B baseline, with especially strong improvements in citation recall from 0.440 to 0.628. The method also attains state-of-the-art results on the official MAGMaR 2026 leaderboard. Code is released at https://github.com/pengyu965/TRACE.

Pretrained large language models (LLMs) are strong in-context learners that are able to perform few-shot learning without changing model parameters. However, as we show, fine-tuning an LLM on any specific task generally destroys its in-context ability. We discover an important cause of this loss, format specialization, where the model overfits to the format of the fine-tuned task and is unable to output anything beyond this format. We further show that format specialization happens at the beginning of fine-tuning. To solve this problem, we propose Prompt Tuning with MOdel Tuning (ProMoT), a simple yet effective two-stage fine-tuning framework that preserves in-context abilities of the pretrained model. ProMoT first trains a soft prompt for the fine-tuning target task, and then fine-tunes the model itself with this soft prompt attached. ProMoT offloads task-specific formats into the soft prompt that can be removed when doing other in-context tasks. We fine-tune mT5 XXL with ProMoT on natural language inference (NLI) and English-French translation and evaluate the in-context abilities of the resulting models on 8 different NLP tasks. ProMoT achieves similar performance on the fine-tuned tasks compared with vanilla fine-tuning, but with much less reduction of in-context learning performances across the board. More importantly, ProMoT shows remarkable generalization ability on tasks that have different formats, e.g. fine-tuning on a NLI binary classification task improves the model's in-context ability to do summarization (+0.53 Rouge-2 score compared to the pretrained model), making ProMoT a promising method to build general purpose capabilities such as grounding and reasoning into LLMs with small but high quality datasets. When extended to sequential or multi-task training, ProMoT can achieve even better out-of-domain generalization performance.

Temporal Grounding (TG) aims to localize video segments corresponding to a textual query. Prior research predominantly focuses on single-segment retrieval. Real-world scenarios, however, often require localizing multiple disjoint segments for a single query -- a setting we term One-to-Many Temporal Grounding (OMTG). Previous state-of-the-art MLLMs, optimized for one-to-one settings, struggle in this context, often yielding near-zero scores due to a lack of event cardinality perception. To bridge this gap, we present a systematic solution with three key contributions. First, we establish the first comprehensive OMTG benchmark, introducing Count Accuracy (C-Acc) and Effective Temporal F1 (EtF1) as evaluation metrics. Second, we curate a high-quality OMTG dataset comprising 56k samples through a sophisticated construction pipeline. Third, we develop novel temporal and caption reward functions specifically designed for OMTG. In particular, the caption reward leverages Chain-of-Thought reasoning over dense video captions to explicitly guide policy optimization toward both preciseness and completeness. Extensive experiments show our model achieves a new state-of-the-art EtF1 of 43.65\% on OMTG Bench, outperforming Gemini 2.5 Pro and Seed-1.8 by 15.85\% and 15.61\%, respectively.

Large Language Models (LLMs) are fluent but prone to hallucinations, producing answers that appear plausible yet are unsupported by available evidence. This failure is especially problematic in high-stakes domains where decisions must be justified by verifiable information. We introduce EvidenceRL, a reinforcement learning framework that enforces evidence adherence during training. EvidenceRL scores candidate responses for grounding (entailment with retrieved evidence and context) and correctness (agreement with reference answers) and optimizes the generator using Group Relative Policy Optimization (GRPO). We evaluate across two high-stakes domains, cardiac diagnosis and legal reasoning, where EvidenceRL consistently improves evidence grounding and faithfulness without sacrificing task accuracy. On cardiac diagnosis, F1@3 increases from 37.0 to 54.5 on Llama-3.2-3B while grounding (G_{max}@3) rises from 47.6 to 78.2; hallucinations drop nearly 5times and evidence-supported diagnoses increase from 31.8\% to 61.6\%. On legal reasoning, EvidenceRL raises Faithfulness from 32.8\% to 67.6\% on Llama-3.1-8B, demonstrating consistent behavioral change across domains. Our code is open-sourced at https://github.com/Wizaaard/EvidenceRL.git.

This paper introduces two significant contributions to address the issue of grounding claims in a given context. Grounding means that given a context (document) and a claim, there's at least one supportive evidence for the claim in the document. We will introduce Paladin-mini, a compact (3.8B parameters) open-source classifier model (used for labeling data as grounded or ungrounded) engineered for robust performance in real-world scenarios, and the grounding-benchmark, a new evaluation dataset designed to assess performance on critical reasoning tasks. We'll also demonstrate the results of Paladin-mini with benchmarks against the current State-of-the-art and share clear and reproducible results.

Super-Resolution (SR) has advanced rapidly in recent years, with diffusion-based models achieving unprecedented fidelity at the cost of introducing new types of visual artifacts. While existing Image Quality Assessment (IQA) methods provide holistic quality scores, they lack interpretability and fail to distinguish between different artifact types arising from modern SR approaches. To address this gap, we introduce SR-Ground, a large-scale dataset specifically designed for fine-grained artifact segmentation in super-resolved images. The dataset comprises images processed by a diverse set of state-of-the-art SR models, with pixel-level annotations for multiple artifact categories. We conduct a large-scale crowdsourcing study involving 1,062 participants to validate and refine automatically generated segmentations, resulting in a high-quality dataset of 63,000 images spanning 6 distinct artifact types. We demonstrate that training IQA models with grounding capabilities on SR-Ground significantly improves performance on downstream tasks. Furthermore, we introduce a fine-tuning pipeline that leverages our grounding model to reduce perceptible artifacts in SR outputs, showcasing the practical utility of our dataset.

Graphical user interface (GUI) grounding is a key function of computer-use agents, which maps natural-language instructions to actionable screen regions. Existing approaches based on Multimodal Large Language Models (MLLMs) typically formulate it as a text-based coordinate generation task, yet directly generating precise coordinates from visual inputs remains challenging and computationally intensive. An intuitive way to implement GUI grounding is to first select visual patches relevant to the instructions and then determine the precise click location within those patches. Based on the observations that general MLLMs have some native grounding capability, nested within their attentions, we propose GUI-AIMA, an attention-based and coordinate-free supervised fine-tuning framework for efficient GUI grounding. GUI-AIMA aligns the intrinsic multimodal attention of MLLMs with patch-wise grounding signals. These signals are calculated adaptively for diverse user instructions by multi-head aggregation on simplified query-visual attention matrices. Besides, its coordinate-free manner can easily integrate a plug-and-play zoom-in stage. GUI-AIMA-3B was trained with only 85k screenshots, demonstrating exceptional data efficiency and verifying that light training can trigger the native grounding capability of MLLMs. It achieves state-of-the-art performance among 3B models, attaining an average accuracy of 58.6% on ScreenSpot-Pro and 62.2% on OSWorld-G. Project page: https://github.com/sjz5202/GUI-AIMA

Graphical user interface (GUI) grounding, the process of mapping human instructions to GUI actions, serves as a fundamental basis to autonomous GUI agents. While existing grounding models achieve promising performance to simulate the mouse click action on various click-based benchmarks, another essential mode of mouse interaction, namely dragging, remains largely underexplored. Yet, dragging the mouse to select and manipulate textual content represents a prevalent and important usage in practical GUI scenarios. To narrow this gap, we first introduce GUI-Drag, a diverse dataset of 161K text dragging examples synthesized through a scalable pipeline. To support systematic and robust evaluation, we further construct ScreenDrag, a benchmark with 5,333 examples spanning three levels of interface context, together with three dedicated metrics designed for assessing text dragging capability. Models trained on GUI-Drag with an efficient continual training strategy achieve substantial improvements on ScreenDrag, while preserving the original click-based performance on ScreenSpot, ScreenSpot-v2, and OSWorld-G. Our work encourages further research on broader GUI grounding beyond just clicking and paves way toward a truly generalist GUI grounding model. All benchmark, data, checkpoints, and code are open-sourced and available at https://osu-nlp-group.github.io/GUI-Drag.

Recent advancements in large language models (LLM) capable of processing extremely long texts highlight the need for a dedicated evaluation benchmark to assess their long-context capabilities. However, existing methods, like the needle-in-a-haystack test, do not effectively assess whether these models fully utilize contextual information, raising concerns about the reliability of current evaluation techniques. To thoroughly examine the effectiveness of existing benchmarks, we introduce a new metric called information coverage (IC), which quantifies the proportion of the input context necessary for answering queries. Our findings indicate that current benchmarks exhibit low IC; although the input context may be extensive, the actual usable context is often limited. To address this, we present ETHIC, a novel benchmark designed to assess LLMs' ability to leverage the entire context. Our benchmark comprises 2,648 test instances spanning four long-context tasks with high IC scores in the domains of books, debates, medicine, and law. Our evaluations reveal significant performance drops in contemporary LLMs, highlighting a critical challenge in managing long contexts. Our benchmark is available at https://github.com/dmis-lab/ETHIC.

Generative retrieval models encode pointers to information in a corpus as an index within the model's parameters. These models serve as part of a larger pipeline, where retrieved information conditions generation for knowledge-intensive NLP tasks. However, we identify two limitations: the generative retrieval does not account for contextual information. Secondly, the retrieval can't be tuned for the downstream readers as decoding the page title is a non-differentiable operation. This paper introduces Re3val, trained with generative reranking and reinforcement learning using limited data. Re3val leverages context acquired via Dense Passage Retrieval to rerank the retrieved page titles and utilizes REINFORCE to maximize rewards generated by constrained decoding. Additionally, we generate questions from our pre-training dataset to mitigate epistemic uncertainty and bridge the domain gap between the pre-training and fine-tuning datasets. Subsequently, we extract and rerank contexts from the KILT database using the rerank page titles. Upon grounding the top five reranked contexts, Re3val demonstrates the Top 1 KILT scores compared to all other generative retrieval models across five KILT datasets.

This paper introduces UI-TARS, a native GUI agent model that solely perceives the screenshots as input and performs human-like interactions (e.g., keyboard and mouse operations). Unlike prevailing agent frameworks that depend on heavily wrapped commercial models (e.g., GPT-4o) with expert-crafted prompts and workflows, UI-TARS is an end-to-end model that outperforms these sophisticated frameworks. Experiments demonstrate its superior performance: UI-TARS achieves SOTA performance in 10+ GUI agent benchmarks evaluating perception, grounding, and GUI task execution. Notably, in the OSWorld benchmark, UI-TARS achieves scores of 24.6 with 50 steps and 22.7 with 15 steps, outperforming Claude (22.0 and 14.9 respectively). In AndroidWorld, UI-TARS achieves 46.6, surpassing GPT-4o (34.5). UI-TARS incorporates several key innovations: (1) Enhanced Perception: leveraging a large-scale dataset of GUI screenshots for context-aware understanding of UI elements and precise captioning; (2) Unified Action Modeling, which standardizes actions into a unified space across platforms and achieves precise grounding and interaction through large-scale action traces; (3) System-2 Reasoning, which incorporates deliberate reasoning into multi-step decision making, involving multiple reasoning patterns such as task decomposition, reflection thinking, milestone recognition, etc. (4) Iterative Training with Reflective Online Traces, which addresses the data bottleneck by automatically collecting, filtering, and reflectively refining new interaction traces on hundreds of virtual machines. Through iterative training and reflection tuning, UI-TARS continuously learns from its mistakes and adapts to unforeseen situations with minimal human intervention. We also analyze the evolution path of GUI agents to guide the further development of this domain.

When provided with sufficient explanatory context, smaller Language Models have been shown to exhibit strong reasoning ability on challenging short-answer question-answering tasks where the questions are unseen in training. We evaluate two methods for further improvement in this setting. Both methods focus on combining rationales generated by a larger Language Model with longer contexts created from a multi-hop dense retrieval system. The first method (RR) involves training a Rationale Ranking model to score both generated rationales and retrieved contexts with respect to relevance and truthfulness. We then use the scores to derive combined contexts from both knowledge sources using a number of combinatory strategies. For the second method (RATD) we utilise retrieval-augmented training datasets developed by Hartill et al. 2023 to train a smaller Reasoning model such that it becomes proficient at utilising relevant information from longer text sequences that may be only partially evidential and frequently contain many irrelevant sentences. We find that both methods significantly improve results. Our single best Reasoning model materially improves upon strong comparable prior baselines for unseen evaluation datasets (StrategyQA 58.9 rightarrow 61.7 acc., CommonsenseQA 63.6 rightarrow 72.7 acc., ARC-DA 31.6 rightarrow 52.1 F1, IIRC 25.5 rightarrow 27.3 F1) and a version utilising our prior knowledge of each type of question in selecting a context combination strategy does even better. Our proposed models also generally outperform direct prompts against much larger models (BLOOM 175B and StableVicuna 13B) in both few-shot chain-of-thought and standard few-shot settings.

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 Video-to-text Information Bottleneck Evaluation (VIBE), an annotation-free method that scores VLM outputs using two metrics: grounding (how well the summary aligns with visual content) and 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 LearningPaper24, SUTD-TrafficQA, and 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.

Context-grounded hallucinations are cases where model outputs contain information not verifiable against the source text. We study the applicability of LLMs for localizing such hallucinations, as a more practical alternative to existing complex evaluation pipelines. In the absence of established benchmarks for meta-evaluation of hallucinations localization, we construct one tailored to LLMs, involving a challenging human annotation of over 1,000 examples. We complement the benchmark with an LLM-based evaluation protocol, verifying its quality in a human evaluation. Since existing representations of hallucinations limit the types of errors that can be expressed, we propose a new representation based on free-form textual descriptions, capturing the full range of possible errors. We conduct a comprehensive study, evaluating four large-scale LLMs, which highlights the benchmark's difficulty, as the best model achieves an F1 score of only 0.67. Through careful analysis, we offer insights into optimal prompting strategies for the task and identify the main factors that make it challenging for LLMs: (1) a tendency to incorrectly flag missing details as inconsistent, despite being instructed to check only facts in the output; and (2) difficulty with outputs containing factually correct information absent from the source - and thus not verifiable - due to alignment with the model's parametric knowledge.

Identifying relevant persona or knowledge for conversational systems is critical to grounded dialogue response generation. However, each grounding has been mostly researched in isolation with more practical multi-context dialogue tasks introduced in recent works. We define Persona and Knowledge Dual Context Identification as the task to identify persona and knowledge jointly for a given dialogue, which could be of elevated importance in complex multi-context dialogue settings. We develop a novel grounding retrieval method that utilizes all contexts of dialogue simultaneously. Our method requires less computational power via utilizing neural QA retrieval models. We further introduce our novel null-positive rank test which measures ranking performance on semantically dissimilar samples (i.e. hard negatives) in relation to data augmentation.

Learning to rank is a key component of many e-commerce search engines. In learning to rank, one is interested in optimising the global ordering of a list of items according to their utility for users.Popular approaches learn a scoring function that scores items individually (i.e. without the context of other items in the list) by optimising a pointwise, pairwise or listwise loss. The list is then sorted in the descending order of the scores. Possible interactions between items present in the same list are taken into account in the training phase at the loss level. However, during inference, items are scored individually, and possible interactions between them are not considered. In this paper, we propose a context-aware neural network model that learns item scores by applying a self-attention mechanism. The relevance of a given item is thus determined in the context of all other items present in the list, both in training and in inference. We empirically demonstrate significant performance gains of self-attention based neural architecture over Multi-LayerPerceptron baselines, in particular on a dataset coming from search logs of a large scale e-commerce marketplace, Allegro.pl. This effect is consistent across popular pointwise, pairwise and listwise losses.Finally, we report new state-of-the-art results on MSLR-WEB30K, the learning to rank benchmark.

Multimodal Large Language Models (MLLMs) are increasingly deployed in human-facing roles where personality perception is critical, yet existing benchmarks evaluate this capability solely on numerical Big Five score prediction, leaving open whether models truly perceive personality through behavioral understanding or merely prejudge through superficial pattern matching. We address this gap with three contributions. (i) A new task: we formalize Grounded Personality Reasoning (GPR), which requires MLLMs to anchor each Big Five rating in observable evidence through a chain of rating, reasoning, and grounding. (ii) A new dataset: we release MM-OCEAN (1,104 videos, 5,320 MCQs), produced by a multi-agent pipeline with human verification, with timestamped behavioral observations, evidence-grounded trait analyses, and seven categories of cue-grounding MCQs. (iii) Benchmark and analysis: we design a three-tier evaluation (rating, reasoning, grounding) plus four sample-level failure-mode metrics: Prejudice Rate (PR), Confabulation Rate (CR), Integration-failure Rate (IR), and Holistic-grounding Rate (HR), and benchmark 27 MLLMs (13 closed, 14 open). The analysis uncovers a striking Prejudice Gap: across the field, 51% of correct ratings are not grounded in retrieved cues, and the Holistic-Grounding Rate spans only 0-33.5%. These findings expose a disconnect between getting the right score and reasoning for the right reason, charting a roadmap for grounded social cognition in MLLMs.

Large language models (LLMs) still produce plausible-sounding but ungrounded factual claims, a problem that worsens in multi-turn dialogue as context grows and early errors cascade. We introduce HalluHard, a challenging multi-turn hallucination benchmark with 950 seed questions spanning four high-stakes domains: legal cases, research questions, medical guidelines, and coding. We operationalize groundedness by requiring inline citations for factual assertions. To support reliable evaluation in open-ended settings, we propose a judging pipeline that iteratively retrieves evidence via web search. It can fetch, filter, and parse full-text sources (including PDFs) to assess whether cited material actually supports the generated content. Across a diverse set of frontier proprietary and open-weight models, hallucinations remain substantial even with web search (approx 30% for the strongest configuration, Opus-4.5 with web search), with content-grounding errors persisting at high rates. Finally, we show that hallucination behavior is shaped by model capacity, turn position, effective reasoning, and the type of knowledge required.

Automatic Speech Scoring (ASS) is the computer-assisted evaluation of a candidate's speaking proficiency in a language. ASS systems face many challenges like open grammar, variable pronunciations, and unstructured or semi-structured content. Recent deep learning approaches have shown some promise in this domain. However, most of these approaches focus on extracting features from a single audio, making them suffer from the lack of speaker-specific context required to model such a complex task. We propose a novel deep learning technique for non-native ASS, called speaker-conditioned hierarchical modeling. In our technique, we take advantage of the fact that oral proficiency tests rate multiple responses for a candidate. We extract context vectors from these responses and feed them as additional speaker-specific context to our network to score a particular response. We compare our technique with strong baselines and find that such modeling improves the model's average performance by 6.92% (maximum = 12.86%, minimum = 4.51%). We further show both quantitative and qualitative insights into the importance of this additional context in solving the problem of ASS.

Composers develop music from an original idea, controlling how similar each new section is to it, from near repetition to new material. Game music needs this often, as its material repeats over long play. Music models, however, rarely learn to control similarity from real pieces. We further train a pretrained MIDI model on section pairs from Super Mario pieces. A label states each pair's similarity, and a context vector summarises the given section, or context. For 157 test contexts, the model generates six outputs per label. Twelve measures score the outputs; one main measure each covers melody, pitch content, rhythm and note length. On average, outputs requested as similar are more similar to their context than those requested as dissimilar, and not to other pieces. All twelve measures favour them; each main measure stays significant after correction for multiple tests. On melody, the gap comes mainly from more frequent restatement of the context's melody. Against an otherwise identical model without a label, only the request for high similarity makes outputs more similar. In a listening test, ratings followed melodic overlap with the context most closely of all measures.

Large language models now solve many benchmark math problems at near-expert levels, yet this progress has not fully translated into reliable performance in real-world applications. We study this gap through contextual mathematical reasoning, where the mathematical core must be formulated from descriptive scenarios. We introduce ContextMATH, a benchmark that repurposes AIME and MATH-500 problems into two contextual settings: Scenario Grounding (SG), which embeds abstract problems into realistic narratives without increasing reasoning complexity, and Complexity Scaling (CS), which transforms explicit conditions into sub-problems to capture how constraints often appear in practice. Evaluating 61 proprietary and open-source models, we observe sharp drops: on average, open-source models decline by 13 and 34 points on SG and CS, while proprietary models drop by 13 and 20. Error analysis shows that errors are dominated by incorrect problem formulation, with formulation accuracy declining as original problem difficulty increases. Correct formulation emerges as a prerequisite for success, and its sufficiency improves with model scale, indicating that larger models advance in both understanding and reasoning. Nevertheless, formulation and reasoning remain two complementary bottlenecks that limit contextual mathematical problem solving. Finally, we find that fine-tuning with scenario data improves performance, whereas formulation-only training is ineffective. However, performance gaps are only partially alleviated, highlighting contextual mathematical reasoning as a central unsolved challenge for LLMs.

Omni LLMs can naturally locate key audio and visual evidence and integrate information across modalities for audiovisual reasoning. Existing audio-visual reasoning benchmarks that mainly rely on multiple-choice questions usually score models solely based on answer correctness, making it difficult to assess whether a model truly forms a reasoning chain from evidence to the correct answer. In addition, textual or unimodal shortcuts may allow models to answer correctly without analyzing the necessary audio-visual evidence, further weakening the reliability of evaluation benchmarks. To address these issues, we introduce EvaOmni, an evidence-grounded benchmark containing 392 videos and 1,027 purely human-annotated question-answer pairs, covering 15 capabilities and three levels: retrieval and localization, basic understanding, and complex reasoning. We design a series of rigorous annotation procedures in which experts annotate audio and visual evidence and construct video-based distractors; humans review whether questions meet strict inspection criteria, and finally models filter textual and unimodal shortcuts. Beyond high-quality question-answer pairs, we require models to simultaneously return the answer and supporting evidence intervals in order to evaluate the complete reasoning chain. We propose the Coverage Grounded Score (C.G.) to measure how much annotated evidence a model identifies, and the Bottleneck Grounded Score (B.G.) to penalize poor temporal localization in either modality; both are incorporated into answer correctness. Across 7 open-weight and 17 proprietary Omni LLMs, Gemini3.7-Flash achieved the highest overall accuracy (79.4%), C.G. (70.8%), and B.G. (35.1%). At Level 3, it reached 79.1% accuracy, but its B.G. was only 12.3%. These results show that even the most advanced models still have room for development in grounded audio-visual reasoning.

Context engineering has become central to building reliable AI agents, yet it remains largely unmeasured. Agents do not fail in isolation: their behavior is shaped by the instructions, tools, memory, retrieved knowledge, guardrails, and untrusted inputs accumulated in their context. When this context is weak, agents drift, hallucinate, misuse tools, ignore constraints, become vulnerable to injection, and waste tokens. This paper validates context-engineering quality as an independent leading indicator of agent reliability. We implement the measurement in ProofAgent-Harness, an open-source infrastructure for AI agent evaluation that uses multi-juror, consensus-based scoring. The harness assesses context across seven criteria: role clarity, guardrail coverage, instruction consistency, tool schema quality, grounding sufficiency, injection hardening, and token efficiency. Crucially, the context score is isolated from behavioral metrics and release decisions, enabling a non-circular validation. Through a controlled context-quality study across regulated agent domains, holding frontier LLM agents fixed and varying only their operating context, we show that context-quality criteria consistently predict their corresponding behavioral outcomes. Grounding sufficiency predicts hallucination resistance, guardrail coverage predicts manipulation resistance, instruction consistency predicts instruction following, and tool-schema quality predicts tool use. These findings establish context measurement as a validated preflight signal for agent reliability and position context engineering as an auditable layer of agent evaluation and governance.

We introduce FACTS Grounding, an online leaderboard and associated benchmark that evaluates language models' ability to generate text that is factually accurate with respect to given context in the user prompt. In our benchmark, each prompt includes a user request and a full document, with a maximum length of 32k tokens, requiring long-form responses. The long-form responses are required to be fully grounded in the provided context document while fulfilling the user request. Models are evaluated using automated judge models in two phases: (1) responses are disqualified if they do not fulfill the user request; (2) they are judged as accurate if the response is fully grounded in the provided document. The automated judge models were comprehensively evaluated against a held-out test-set to pick the best prompt template, and the final factuality score is an aggregate of multiple judge models to mitigate evaluation bias. The FACTS Grounding leaderboard will be actively maintained over time, and contains both public and private splits to allow for external participation while guarding the integrity of the leaderboard. It can be found at https://www.kaggle.com/facts-leaderboard.

As LLM benchmarks saturate, the evaluation community has pursued two strategies to increase difficulty: escalating knowledge demands (GPQA, HLE) or removing knowledge entirely in favor of abstract reasoning (ARC-AGI). The first conflates memorization with capability; the second divorces reasoning from the practical contexts in which it matters. We take a different approach. The Grounded Integration Measure (GIM) is a benchmark of 820 original problems (615 public, 205 private) where difficulty comes from integration; individual problems require coordinating multiple cognitive operations (constraint satisfaction, state tracking, epistemic vigilance, audience calibration) over broadly accessible knowledge, so that reasoning stays grounded in realistic tasks without being gated on specialized expertise. Each problem is an original expert-authored composition, majority with rubric-decomposed scoring (median 6 independently judged criteria). A balanced public--private split provides built-in contamination diagnostic. We calibrate a continuous response 2-parameter logistic (2PL) IRT model over >200k prompt-response pairs across 28 models, producing robust ability estimates that correctly order test-configurations even when raw accuracy is distorted by errors or missing data, addressing a common challenge in benchmark reporting. Using this framework, we present a comprehensive leaderboard spanning 22 models and 47 test-configurations (unique model, thinking-level pairs), and conduct what is to our knowledge the most extensive published study of how test-time compute trades off against model capability on a fixed benchmark: 11 models swept across 35 test-configurations. We observe that within-family configuration choices, such as thinking budget and quantization, matter as much as model selection. We release the evaluation framework, calibrated IRT parameters, and all public problems.

Recent large language models (LLMs) support long contexts ranging from 128K to 1M tokens. A popular method for evaluating these capabilities is the needle-in-a-haystack (NIAH) test, which involves retrieving a "needle" (relevant information) from a "haystack" (long irrelevant context). Extensions of this approach include increasing distractors, fact chaining, and in-context reasoning. However, in these benchmarks, models can exploit existing literal matches between the needle and haystack to simplify the task. To address this, we introduce NoLiMa, a benchmark extending NIAH with a carefully designed needle set, where questions and needles have minimal lexical overlap, requiring models to infer latent associations to locate the needle within the haystack. We evaluate 12 popular LLMs that claim to support contexts of at least 128K tokens. While they perform well in short contexts (<1K), performance degrades significantly as context length increases. At 32K, for instance, 10 models drop below 50% of their strong short-length baselines. Even GPT-4o, one of the top-performing exceptions, experiences a reduction from an almost-perfect baseline of 99.3% to 69.7%. Our analysis suggests these declines stem from the increased difficulty the attention mechanism faces in longer contexts when literal matches are absent, making it harder to retrieve relevant information.

Forecasting is a critical task in decision-making across numerous domains. While historical numerical data provide a start, they fail to convey the complete context for reliable and accurate predictions. Human forecasters frequently rely on additional information, such as background knowledge and constraints, which can efficiently be communicated through natural language. However, in spite of recent progress with LLM-based forecasters, their ability to effectively integrate this textual information remains an open question. To address this, we introduce "Context is Key" (CiK), a time-series forecasting benchmark that pairs numerical data with diverse types of carefully crafted textual context, requiring models to integrate both modalities; crucially, every task in CiK requires understanding textual context to be solved successfully. We evaluate a range of approaches, including statistical models, time series foundation models, and LLM-based forecasters, and propose a simple yet effective LLM prompting method that outperforms all other tested methods on our benchmark. Our experiments highlight the importance of incorporating contextual information, demonstrate surprising performance when using LLM-based forecasting models, and also reveal some of their critical shortcomings. This benchmark aims to advance multimodal forecasting by promoting models that are both accurate and accessible to decision-makers with varied technical expertise. The benchmark can be visualized at https://servicenow.github.io/context-is-key-forecasting/v0/.

Vision-language models can connect the text description of an object to its specific location in an image through visual grounding. This has potential applications in enhanced radiology reporting. However, these models require large annotated image-text datasets, which are lacking for PET/CT. We developed an automated pipeline to generate weak labels linking PET/CT report descriptions to their image locations and used it to train a 3D vision-language visual grounding model. Our pipeline finds positive findings in PET/CT reports by identifying mentions of SUVmax and axial slice numbers. From 25,578 PET/CT exams, we extracted 11,356 sentence-label pairs. Using this data, we trained ConTEXTual Net 3D, which integrates text embeddings from a large language model with a 3D nnU-Net via token-level cross-attention. The model's performance was compared against LLMSeg, a 2.5D version of ConTEXTual Net, and two nuclear medicine physicians. The weak-labeling pipeline accurately identified lesion locations in 98% of cases (246/251), with 7.5% requiring boundary adjustments. ConTEXTual Net 3D achieved an F1 score of 0.80, outperforming LLMSeg (F1=0.22) and the 2.5D model (F1=0.53), though it underperformed both physicians (F1=0.94 and 0.91). The model achieved better performance on FDG (F1=0.78) and DCFPyL (F1=0.75) exams, while performance dropped on DOTATE (F1=0.58) and Fluciclovine (F1=0.66). The model performed consistently across lesion sizes but showed reduced accuracy on lesions with low uptake. Our novel weak labeling pipeline accurately produced an annotated dataset of PET/CT image-text pairs, facilitating the development of 3D visual grounding models. ConTEXTual Net 3D significantly outperformed other models but fell short of the performance of nuclear medicine physicians. Our study suggests that even larger datasets may be needed to close this performance gap.

Large language models (LLMs) are often supplemented with external knowledge to provide information not encoded in their parameters or to reduce hallucination. In such cases, we expect the model to generate responses by grounding its response in the provided external context. However, prior work has shown that simply appending context at inference time does not ensure grounded generation. To address this, we propose Context-INformed Grounding Supervision (CINGS), a post-training supervision in which the model is trained with relevant context prepended to the response, while computing the loss only over the response tokens and masking out the context. Our experiments demonstrate that models trained with CINGS exhibit stronger grounding in both textual and visual domains compared to standard instruction-tuned models. In the text domain, CINGS outperforms other training methods across 11 information-seeking datasets and is complementary to inference-time grounding techniques. In the vision-language domain, replacing a vision-language model's LLM backbone with a CINGS-trained model reduces hallucinations across four benchmarks and maintains factual consistency throughout the generated response. This improved grounding comes without degradation in general downstream performance. Finally, we analyze the mechanism underlying the enhanced grounding in CINGS and find that it induces a shift in the model's prior knowledge and behavior, implicitly encouraging greater reliance on the external context.

To answer a question, language models often need to integrate prior knowledge learned during pretraining and new information presented in context. We hypothesize that models perform this integration in a predictable way across different questions and contexts: models will rely more on prior knowledge for questions about entities (e.g., persons, places, etc.) that they are more familiar with due to higher exposure in the training corpus, and be more easily persuaded by some contexts than others. To formalize this problem, we propose two mutual information-based metrics to measure a model's dependency on a context and on its prior about an entity: first, the persuasion score of a given context represents how much a model depends on the context in its decision, and second, the susceptibility score of a given entity represents how much the model can be swayed away from its original answer distribution about an entity. Following well-established measurement modeling methods, we empirically test for the validity and reliability of these metrics. Finally, we explore and find a relationship between the scores and the model's expected familiarity with an entity, and provide two use cases to illustrate their benefits.

We develop a set of methods to improve on the results of self-supervised learning using context. We start with a baseline of patch based arrangement context learning and go from there. Our methods address some overt problems such as chromatic aberration as well as other potential problems such as spatial skew and mid-level feature neglect. We prevent problems with testing generalization on common self-supervised benchmark tests by using different datasets during our development. The results of our methods combined yield top scores on all standard self-supervised benchmarks, including classification and detection on PASCAL VOC 2007, segmentation on PASCAL VOC 2012, and "linear tests" on the ImageNet and CSAIL Places datasets. We obtain an improvement over our baseline method of between 4.0 to 7.1 percentage points on transfer learning classification tests. We also show results on different standard network architectures to demonstrate generalization as well as portability. All data, models and programs are available at: https://gdo-datasci.llnl.gov/selfsupervised/.

The development of multimodal large language models (MLLMs) enables the evaluation of image quality through natural language descriptions. This advancement allows for more detailed assessments. However, these MLLM-based IQA methods primarily rely on general contextual descriptions, sometimes limiting fine-grained quality assessment. To address this limitation, we introduce a new image quality assessment (IQA) task paradigm, grounding-IQA. This paradigm integrates multimodal referring and grounding with IQA to realize more fine-grained quality perception. Specifically, grounding-IQA comprises two subtasks: grounding-IQA-description (GIQA-DES) and visual question answering (GIQA-VQA). GIQA-DES involves detailed descriptions with precise locations (e.g., bounding boxes), while GIQA-VQA focuses on quality QA for local regions. To realize grounding-IQA, we construct a corresponding dataset, GIQA-160K, through our proposed automated annotation pipeline. Furthermore, we develop a well-designed benchmark, GIQA-Bench. The benchmark comprehensively evaluates the model grounding-IQA performance from three perspectives: description quality, VQA accuracy, and grounding precision. Experiments demonstrate that our proposed task paradigm, dataset, and benchmark facilitate the more fine-grained IQA application. Code: https://github.com/zhengchen1999/Grounding-IQA.

Understanding context is key to understanding human language, an ability which Large Language Models (LLMs) have been increasingly seen to demonstrate to an impressive extent. However, though the evaluation of LLMs encompasses various domains within the realm of Natural Language Processing, limited attention has been paid to probing their linguistic capability of understanding contextual features. This paper introduces a context understanding benchmark by adapting existing datasets to suit the evaluation of generative models. This benchmark comprises of four distinct tasks and nine datasets, all featuring prompts designed to assess the models' ability to understand context. First, we evaluate the performance of LLMs under the in-context learning pretraining scenario. Experimental results indicate that pre-trained dense models struggle with understanding more nuanced contextual features when compared to state-of-the-art fine-tuned models. Second, as LLM compression holds growing significance in both research and real-world applications, we assess the context understanding of quantized models under in-context-learning settings. We find that 3-bit post-training quantization leads to varying degrees of performance reduction on our benchmark. We conduct an extensive analysis of these scenarios to substantiate our experimental results.

Prior work uses linear probes on benchmark prompts as evidence of evaluation awareness in large language models. Because evaluation context is typically entangled with benchmark format and genre, it is unclear whether probe-based signals reflect context or surface structure. We test whether these signals persist under partial control of prompt format using a controlled 2x2 dataset and diagnostic rewrites. We find that probes primarily track benchmark-canonical structure and fail to generalize to free-form prompts independent of linguistic style. Thus, standard probe-based methodologies do not reliably disentangle evaluation context from structural artifacts, limiting the evidential strength of existing results.

How does retrieval performance scale with pretraining FLOPs? We benchmark retrieval performance across LLM model sizes from 125 million parameters to 7 billion parameters pretrained on datasets ranging from 1 billion tokens to more than 2 trillion tokens. We find that retrieval performance on zero-shot BEIR tasks predictably scales with LLM size, training duration, and estimated FLOPs. We also show that In-Context Learning scores are strongly correlated with retrieval scores across retrieval tasks. Finally, we highlight the implications this has for the development of LLM-based retrievers.

In-context learning (ICL) is a powerful paradigm emerged from large language models (LLMs). Despite its promises, ICL performance is known to be highly sensitive to input examples. In this work, we use in-context influences to analyze few-shot ICL performance directly from the in-context examples. Our proposed influence-based example selection method can identify both positive and negative examples, outperforming several baselines when evaluated on 9 SuperGLUE tasks. Our analysis uncovers up to a 16.3% performance gap between using the most negative in-context examples compared to the most positive. In a case study, we apply our influence-based framework to quantify the phenomena of recency bias in example ordering for few-shot ICL.

The large language model (LLM)-as-judge paradigm has been used to meet the demand for a cheap, reliable, and fast evaluation of model outputs during AI system development and post-deployment monitoring. While judge models -- LLMs finetuned to specialize in assessing and critiquing model outputs -- have been touted as general purpose evaluators, they are typically evaluated only on non-contextual scenarios, such as instruction following. The omission of contextual settings -- those where external information is used as context to generate an output -- is surprising given the increasing prevalence of retrieval-augmented generation (RAG) and summarization use cases. Contextual assessment is uniquely challenging, as evaluation often depends on practitioner priorities, leading to conditional evaluation criteria (e.g., comparing responses based on factuality and then considering completeness if they are equally factual). To address the gap, we propose ContextualJudgeBench, a judge benchmark with 2,000 challenging response pairs across eight splits inspired by real-world contextual evaluation scenarios. We build our benchmark with a multi-pronged data construction pipeline that leverages both existing human annotations and model-based perturbations. Our comprehensive study across 11 judge models and 9 general purpose models, reveals that the contextual information and its assessment criteria present a significant challenge to even state-of-the-art models. For example, OpenAI's o1, the best-performing model, barely reaches 55% consistent accuracy.

Question Answering (QA) tasks, which involve extracting answers from a given context, are relatively straightforward for modern Large Language Models (LLMs) when the context is short. However, long contexts pose challenges due to the quadratic complexity of the self-attention mechanism. This challenge is compounded in Indic languages, which are often low-resource. This study explores context-shortening techniques, including Open Information Extraction (OIE), coreference resolution, Answer Paragraph Selection (APS), and their combinations, to improve QA performance. Compared to the baseline of unshortened (long) contexts, our experiments on four Indic languages (Hindi, Tamil, Telugu, and Urdu) demonstrate that context-shortening techniques yield an average improvement of 4\% in semantic scores and 47\% in token-level scores when evaluated on three popular LLMs without fine-tuning. Furthermore, with fine-tuning, we achieve an average increase of 2\% in both semantic and token-level scores. Additionally, context-shortening reduces computational overhead. Explainability techniques like LIME and SHAP reveal that when the APS model confidently identifies the paragraph containing the answer, nearly all tokens within the selected text receive high relevance scores. However, the study also highlights the limitations of LLM-based QA systems in addressing non-factoid questions, particularly those requiring reasoning or debate. Moreover, verbalizing OIE-generated triples does not enhance system performance. These findings emphasize the potential of context-shortening techniques to improve the efficiency and effectiveness of LLM-based QA systems, especially for low-resource languages. The source code and resources are available at https://github.com/ritwikmishra/IndicGenQA.

Multi-hop reasoning, which requires multi-step reasoning based on the supporting documents within a given context, remains challenging for large language models (LLMs). LLMs often struggle to filter out irrelevant documents within the context, and their performance is sensitive to the absolute position of supporting documents within that context. In this paper, we identify an additional challenge: LLMs' performance is also sensitive to the order, relative position, in which the supporting documents are presented. We refer to this as the misordered context problem. To address this issue, based on the theoretical approach, we propose a simple yet effective method called context repetition (CoRe), which involves prompting the model by repeatedly presenting the context. This ensures that certain contiguous reasoning segments within supporting documents are presented in the optimal order, effectively guiding the model's reasoning in the appropriate direction. Applying CoRe, we improve the F1 score by up to 30%p on multi-hop QA tasks and increase accuracy by up to 70%p on a synthetic task. Additionally, CoRe helps mitigate the well-known "lost-in-the-middle" problem in LLMs and can be effectively combined with retrieval-based approaches utilizing Chain-of-Thought (CoT) reasoning.

Most large language models are trained on linguistic input alone, yet humans appear to ground their understanding of words in sensorimotor experience. A natural solution is to augment LM representations with human judgments of a word's sensorimotor associations (e.g., the Lancaster Sensorimotor Norms), but this raises another challenge: most words are ambiguous, and judgments of words in isolation fail to account for this multiplicity of meaning (e.g., "wooden table" vs. "data table"). We attempted to address this problem by building a new lexical resource of contextualized sensorimotor judgments for 112 English words, each rated in four different contexts (448 sentences total). We show that these ratings encode overlapping but distinct information from the Lancaster Sensorimotor Norms, and that they also predict other measures of interest (e.g., relatedness), above and beyond measures derived from BERT. Beyond shedding light on theoretical questions, we suggest that these ratings could be of use as a "challenge set" for researchers building grounded language models.

Visual grounding, a task to ground (i.e., localize) natural language in images, essentially requires composite visual reasoning. However, existing methods over-simplify the composite nature of language into a monolithic sentence embedding or a coarse composition of subject-predicate-object triplet. In this paper, we propose to ground natural language in an intuitive, explainable, and composite fashion as it should be. In particular, we develop a novel modular network called Neural Module Tree network (NMTree) that regularizes the visual grounding along the dependency parsing tree of the sentence, where each node is a neural module that calculates visual attention according to its linguistic feature, and the grounding score is accumulated in a bottom-up direction where as needed. NMTree disentangles the visual grounding from the composite reasoning, allowing the former to only focus on primitive and easy-to-generalize patterns. To reduce the impact of parsing errors, we train the modules and their assembly end-to-end by using the Gumbel-Softmax approximation and its straight-through gradient estimator, accounting for the discrete nature of module assembly. Overall, the proposed NMTree consistently outperforms the state-of-the-arts on several benchmarks. Qualitative results show explainable grounding score calculation in great detail.

Affordance grounding refers to the task of finding the area of an object with which one can interact. It is a fundamental but challenging task, as a successful solution requires the comprehensive understanding of a scene in multiple aspects including detection, localization, and recognition of objects with their parts, of geo-spatial configuration/layout of the scene, of 3D shapes and physics, as well as of the functionality and potential interaction of the objects and humans. Much of the knowledge is hidden and beyond the image content with the supervised labels from a limited training set. In this paper, we make an attempt to improve the generalization capability of the current affordance grounding by taking the advantage of the rich world, abstract, and human-object-interaction knowledge from pretrained large-scale vision language models. Under the AGD20K benchmark, our proposed model demonstrates a significant performance gain over the competing methods for in-the-wild object affordance grounding. We further demonstrate it can ground affordance for objects from random Internet images, even if both objects and actions are unseen during training. Project site: https://jasonqsy.github.io/AffordanceLLM/

The accuracy frontier of speech-to-text systems has plateaued on academic benchmarks.1 In contrast, industrial benchmarks and adoption in high-stakes domains suggest otherwise. We hypothesize that the primary difference between the two is contextual conditioning: Academic benchmarks are dominated by frequently encountered general vocabulary that is relatively easy to recognize compared with rare and context-defined custom vocabulary that has disproportionate impact on the usability of speech transcripts. Despite progress on contextual speech-to-text, there is no standardized benchmark. We introduce Contextual Earnings-22, an open dataset built upon Earnings-22, with realistic custom vocabulary contexts to foster research and reveal latent progress. We set six strong baselines for two dominant approaches: keyword prompting and keyword boosting. Experiments show both reach comparable and significantly improved accuracy when scaled from proof-of-concept to large-scale systems.

Visual grounding refers to the ability of a model to identify a region within some visual input that matches a textual description. Consequently, a model equipped with visual grounding capabilities can target a wide range of applications in various domains, including referring expression comprehension, answering questions pertinent to fine-grained details in images or videos, caption visual context by explicitly referring to entities, as well as low and high-level control in simulated and real environments. In this survey paper, we review representative works across the key areas of research on modern general-purpose vision language models (VLMs). We first outline the importance of grounding in VLMs, then delineate the core components of the contemporary paradigm for developing grounded models, and examine their practical applications, including benchmarks and evaluation metrics for grounded multimodal generation. We also discuss the multifaceted interrelations among visual grounding, multimodal chain-of-thought, and reasoning in VLMs. Finally, we analyse the challenges inherent to visual grounding and suggest promising directions for future research.

Visual grounding is essential for precise perception and reasoning in multimodal large language models (MLLMs), especially in medical imaging domains. While existing medical visual grounding benchmarks primarily focus on single-image scenarios, real-world clinical applications often involve sequential images, where accurate lesion localization across different modalities and temporal tracking of disease progression (e.g., pre- vs. post-treatment comparison) require fine-grained cross-image semantic alignment and context-aware reasoning. To remedy the underrepresentation of image sequences in existing medical visual grounding benchmarks, we propose MedSG-Bench, the first benchmark tailored for Medical Image Sequences Grounding. It comprises eight VQA-style tasks, formulated into two paradigms of the grounding tasks, including 1) Image Difference Grounding, which focuses on detecting change regions across images, and 2) Image Consistency Grounding, which emphasizes detection of consistent or shared semantics across sequential images. MedSG-Bench covers 76 public datasets, 10 medical imaging modalities, and a wide spectrum of anatomical structures and diseases, totaling 9,630 question-answer pairs. We benchmark both general-purpose MLLMs (e.g., Qwen2.5-VL) and medical-domain specialized MLLMs (e.g., HuatuoGPT-vision), observing that even the advanced models exhibit substantial limitations in medical sequential grounding tasks. To advance this field, we construct MedSG-188K, a large-scale instruction-tuning dataset tailored for sequential visual grounding, and further develop MedSeq-Grounder, an MLLM designed to facilitate future research on fine-grained understanding across medical sequential images. The benchmark, dataset, and model are available at https://huggingface.co/MedSG-Bench

Integrating automatic speech scoring/assessment systems has become a critical aspect of second-language speaking education. With self-supervised learning advancements, end-to-end speech scoring approaches have exhibited promising results. However, this study highlights the significant decrease in the performance of speech scoring systems in new question contexts, thereby identifying this as a cold start problem in terms of items. With the finding of cold-start phenomena, this paper seeks to alleviate the problem by following methods: 1) prompt embeddings, 2) question context embeddings using BERT or CLIP models, and 3) choice of the pretrained acoustic model. Experiments are conducted on TOEIC speaking test datasets collected from English-as-a-second-language (ESL) learners rated by professional TOEIC speaking evaluators. The results demonstrate that the proposed framework not only exhibits robustness in a cold-start environment but also outperforms the baselines for known content.

We describe our system for LitTraceQA (GroundLM @ EMNLP 2026): given a research question, retrieve the relevant papers from a pool of 27,487, cite the page and the table or figure where the answer lives, and answer in a requested format. Our main finding is that evidence grounding is limited by retrieval, not by reading. The page selector put the annotator's page, which we call the gold page, in front of the model that locates evidence only about half the time (52.6% gold-page recall), while that model, given the page, cited the right one in 45 of the 48 locators it emitted (94%). When the page was missing it rarely said so: of 45 such cases it returned nothing 14 times, a wrong page 24 times, and a correct page 7 times, so the pipeline failed quietly almost twice as often as it failed visibly. Since the failure was that the right page was never shown, the fix is to stop choosing: each retrieved paper fits in the model's context, so we show it whole. Page ranking survives only as a fallback inside papers too long to fit, which no test-split paper was, and gold-page recall reaches 100% on the papers we can parse. Separately, questions that identify their target by position rather than content, such as "the first author of the 24th reference", are served by parsing rather than retrieval: we resolve the bibliography into an addressable list, which also supplies identifiers the evidence metric scores. The final system scores 0.762 paper F_1, 0.441 evidence F_1 and 0.920 multiple-choice accuracy on the held-out test split. Because the pipeline depends on a closed model without seed control, we release a harness that verifies the paper's central claims against committed artifacts.

Answer grounding is the task of locating relevant visual evidence for the Visual Question Answering task. While a wide variety of attention methods have been introduced for this task, they suffer from the following three problems: designs that do not allow the usage of pre-trained networks and do not benefit from large data pre-training, custom designs that are not based on well-grounded previous designs, therefore limiting the learning power of the network, or complicated designs that make it challenging to re-implement or improve them. In this paper, we propose a novel architectural block, which we term Sentence Attention Block, to solve these problems. The proposed block re-calibrates channel-wise image feature-maps by explicitly modeling inter-dependencies between the image feature-maps and sentence embedding. We visually demonstrate how this block filters out irrelevant feature-maps channels based on sentence embedding. We start our design with a well-known attention method, and by making minor modifications, we improve the results to achieve state-of-the-art accuracy. The flexibility of our method makes it easy to use different pre-trained backbone networks, and its simplicity makes it easy to understand and be re-implemented. We demonstrate the effectiveness of our method on the TextVQA-X, VQS, VQA-X, and VizWiz-VQA-Grounding datasets. We perform multiple ablation studies to show the effectiveness of our design choices.

With the development of multimodal reasoning models, Computer Use Agents (CUAs), akin to Jarvis from "Iron Man", are becoming a reality. GUI grounding is a core component for CUAs to execute actual actions, similar to mechanical control in robotics, and it directly leads to the success or failure of the system. It determines actions such as clicking and typing, as well as related parameters like the coordinates for clicks. Current end-to-end grounding models still achieve less than 65\% accuracy on challenging benchmarks like ScreenSpot-pro and UI-Vision, indicating they are far from being ready for deployment. % , as a single misclick can result in unacceptable consequences. In this work, we conduct an empirical study on the training of grounding models, examining details from data collection to model training. Ultimately, we developed the Phi-Ground model family, which achieves state-of-the-art performance across all five grounding benchmarks for models under 10B parameters in agent settings. In the end-to-end model setting, our model still achieves SOTA results with scores of \textbf{43.2} on ScreenSpot-pro and \textbf{27.2} on UI-Vision. We believe that the various details discussed in this paper, along with our successes and failures, not only clarify the construction of grounding models but also benefit other perception tasks. Project homepage: https://zhangmiaosen2000.github.io/Phi-Ground/{https://zhangmiaosen2000.github.io/Phi-Ground/}

Long-context models (LCMs) have demonstrated great potential in processing long sequences, facilitating many real-world applications. The success of LCMs can be attributed to their ability to locate implicit critical information within the context for further prediction. However, recent research reveals that LCMs are often susceptible to contextual noise, i.e., irrelevant tokens, that can mislead model attention. In this paper, we conduct a fine-grained analysis of the context noise and propose an effective metric, the Integrated Gradient (IG) score, to detect and quantify the noise information within the context. Our findings reveal that even simple mitigation of detected context noise can substantially boost the model's attention on critical tokens and benefit subsequent predictions. Building on this insight, we propose Context Denoising Training (CDT), a straightforward yet effective training strategy that improves attention on critical tokens while reinforcing their influence on model predictions. Extensive experiments across four tasks, under both context window scaling and long-context alignment settings, demonstrate the superiority of CDT. Notably, when trained with CDT, an open-source 8B model can achieve performance (50.92) comparable to GPT-4o (51.00).

Grounding (i.e. localizing) arbitrary, free-form textual phrases in visual content is a challenging problem with many applications for human-computer interaction and image-text reference resolution. Few datasets provide the ground truth spatial localization of phrases, thus it is desirable to learn from data with no or little grounding supervision. We propose a novel approach which learns grounding by reconstructing a given phrase using an attention mechanism, which can be either latent or optimized directly. During training our approach encodes the phrase using a recurrent network language model and then learns to attend to the relevant image region in order to reconstruct the input phrase. At test time, the correct attention, i.e., the grounding, is evaluated. If grounding supervision is available it can be directly applied via a loss over the attention mechanism. We demonstrate the effectiveness of our approach on the Flickr 30k Entities and ReferItGame datasets with different levels of supervision, ranging from no supervision over partial supervision to full supervision. Our supervised variant improves by a large margin over the state-of-the-art on both datasets.

In-context learning (ICL) has shown impressive results in few-shot learning tasks, yet its underlying mechanism is still not fully understood. A recent line of work suggests that ICL performs gradient descent (GD)-based optimization implicitly. While appealing, much of the research focuses on simplified settings, where the parameters of a shallow model are optimized. In this work, we revisit evidence for ICL-GD correspondence on realistic NLP tasks and models. We find gaps in evaluation, both in terms of problematic metrics and insufficient baselines. We show that surprisingly, even untrained models achieve comparable ICL-GD similarity scores despite not exhibiting ICL. Next, we explore a major discrepancy in the flow of information throughout the model between ICL and GD, which we term Layer Causality. We propose a simple GD-based optimization procedure that respects layer causality, and show it improves similarity scores significantly.

Visual grounding, localizing objects from natural language descriptions, represents a critical bridge between language and vision understanding. While multimodal large language models (MLLMs) achieve impressive scores on existing benchmarks, a fundamental question remains: can MLLMs truly ground language in vision with human-like sophistication, or are they merely pattern-matching on simplified datasets? Current benchmarks fail to capture real-world complexity where humans effortlessly navigate ambiguous references and recognize when grounding is impossible. To rigorously assess MLLMs' true capabilities, we introduce GroundingME, a benchmark that systematically challenges models across four critical dimensions: (1) Discriminative, distinguishing highly similar objects, (2) Spatial, understanding complex relational descriptions, (3) Limited, handling occlusions or tiny objects, and (4) Rejection, recognizing ungroundable queries. Through careful curation combining automated generation with human verification, we create 1,005 challenging examples mirroring real-world complexity. Evaluating 25 state-of-the-art MLLMs reveals a profound capability gap: the best model achieves only 45.1% accuracy, while most score 0% on rejection tasks, reflexively hallucinating objects rather than acknowledging their absence, raising critical safety concerns for deployment. We explore two strategies for improvements: (1) test-time scaling selects optimal response by thinking trajectory to improve complex grounding by up to 2.9%, and (2) data-mixture training teaches models to recognize ungroundable queries, boosting rejection accuracy from 0% to 27.9%. GroundingME thus serves as both a diagnostic tool revealing current limitations in MLLMs and a roadmap toward human-level visual grounding.

Before letting an agent operate over real context, can you prove it used the right evidence? GroundEval turns that question into a deterministic test of what the agent searched, fetched, cited, and was permitted to access. In one case study, two frontier LLM judges scored a plausible agent response above 0.85. But the trace told a different story: the agent had never retrieved the artifact its answer depended on, yielding a GroundEval score of 0.000. We introduce GroundEval, a judge-free framework for evaluating agents against grounded, time-bounded, and access-controlled evidence. GroundEval uses a domain configuration to generate questions, lets the agent choose how to answer, and then scores both the final answer and the recorded trajectory that produced it. The benchmark targets three failures that LLM-as-judge evaluation struggles to detect: whether an agent checked before claiming absence, reasoned only from evidence available to the actor at the relevant time, and used the correct causal mechanism rather than a plausible one. These correspond to three tracks: Silence, Perspective, and Counterfactual. GroundEval exposes when plausible answers rest on invalid evidence paths, and produces structured per-question diagnostics that pair tool activity with the agent's turn-level narration, making each score inspectable rather than merely reported. What our case studies turned up is that this gap isn't some rare corner case. It's exactly the blind spot that final-answer and judge-based scoring were never built to catch.

Large language models are increasingly capable at closed-world mathematical reasoning, but research assistance also requires source-grounded use of the literature. When a proof reaches a non-trivial step, a useful assistant should determine whether the needed tool (e.g., a lemma) already exists, identify a suitable scholarly source, and verify that its assumptions align with the current proof context. To rigorously evaluate such capabilities, we introduce Re^2Math, a benchmark for tool-grounded retrieval from partial mathematical proofs. Each instance is built from a candidate instrumental citation in the proof of a main theorem, with hierarchical context and an optional leakage-controlled anchor hint. We also make the task source-grounded yet citation-agnostic in that any admissible theorem sufficient for the proof transition is accepted. Evaluation uses a release-frozen retrieval artifact, ensuring reproducibility, while the benchmark itself supports automatic, continual expansion with newly constructed instances. On the current benchmark test set, the best fixed-judge ToolAcc reaches 7.0%, despite substantially higher rates of source grounding, indicating that current systems often retrieve valid statements but fail to establish their applicability to the local proof step. By decoupling citation recall, grounding, and proof-gap sufficiency, Re^2Math transforms literature-grounded mathematical tool use into a controlled diagnostic task.

Tabular foundation models perform in-context learning (ICL) by conditioning predictions on labeled training examples provided as context. Unlike traditional models that separate training from inference, these models must process all training examples in every forward pass, making each prediction expensive. Restricting the number of training examples reduces this cost but substantially degrades performance. Instead of discarding context, we propose activation alignment, a method that leverages the full context to teach a model how to behave when seeing only a subset. This is achieved by training a lightweight linear transformation on synthetic unlabeled data to map the intermediate activations of a data-constrained "student" (using partial context) toward those of a full-context "teacher" (using all data). Training the aligner requires no GPU and converges in seconds to minutes on commodity hardware. We evaluate on 38 classification datasets from the TabArena benchmark using the leading two tabular foundation models, TabPFN-3 and TabFM. Across all context budgets, the aligned student yields broad, statistically significant improvements over the unaligned baseline for both models. In low-data regimes, alignment recovers nearly half of the teacher's predictive advantage. The method provides a practical, low-overhead approach to achieving the inference speed of compact contexts while closing a significant fraction of the performance gap to the full-context teacher.

Automatic Speech Recognition (ASR) has been extensively investigated, yet prior evaluative efforts have largely been restricted to contextless paradigms. This constraint stems from the limited proficiency of conventional ASR models in context modeling and their deficiency in memory and reasoning based on world knowledge. Recent breakthroughs in the development of Large Language Models (LLMs) and corresponding Large Audio Language Models (LALMs) have markedly enhanced the visibility of general artificial intelligence capabilities. Consequently, there exists a compelling need for a benchmark that can evaluate both the generality and intelligence of ASR systems. To address this gap, we propose ContextASR-Bench: a comprehensive, large-scale benchmark designed to assess contextual speech recognition. This benchmark encompasses up to 40,000 data entries across over 10 domains, enabling a thorough evaluation of model performance in scenarios that omit or incorporate coarse-grained or fine-grained contextual information. Moreover, diverging from conventional ASR evaluations, our benchmark includes an analysis of model efficacy in recognizing named entities mentioned within the auditory input. Our extensive evaluation highlights that LALMs, with strong world knowledge and context learning capabilities, outperform conventional ASR models by a large margin. The dataset and evaluation code have been released at https://github.com/MrSupW/ContextASR-Bench.

Large language models can still hallucinate in retrieval-augmented generation (RAG), producing claims that are unsupported by or conflict with the retrieved context. Detecting such errors remains challenging when faithfulness is judged solely against the retrieved context: many existing detectors return holistic answer-level scores, while others target open-domain factuality or fail to provide evidence-grounded diagnostics. We present RT4CHART, a retromorphic testing framework for context-faithfulness assessment. RT4CHART decomposes an answer into independently verifiable claims, performs hierarchical local-to-global verification against the retrieved context, and assigns each claim one of three labels: entailed, contradicted, or baseless. It further maps these claim-level decisions back to specific answer spans and returns explicit context-side evidence, enabling fine-grained auditing rather than opaque scoring. We evaluate RT4CHART on RAGTruth++ (408 samples) and our re-annotated RAGTruth-Enhance (2,675 samples). RT4CHART achieves the best answer-level hallucination-detection F1 score among the evaluated baselines. On RAGTruth++, it attains a precision of 0.845, a recall of 0.718, and an F1 score of 0.776, representing an 83% relative improvement over the strongest baseline. It also achieves a span-level F1 score of 47.5% on RAGTruth-Enhance. Ablation studies show that claim-based local processing drives most of the observed improvement, while global verification provides selective benefits across datasets. Finally, our re-annotation identifies 1.68X more hallucination cases than the original labels, suggesting that commonly used benchmarks substantially underestimate the prevalence of hallucination.

In-context learning (ICL) allows a language model to improve its problem-solving capability when provided with suitable information in context. Since the choice of in-context information can be determined based on the problem itself, in-context learning is analogous to human learning from teachers in a classroom. Recent works (Didolkar et al., 2024a; 2024b) show that ICL performance can be improved by leveraging a frontier large language model's (LLM) ability to predict required skills to solve a problem, popularly referred to as an LLM's metacognition, and using the recommended skills to construct necessary in-context examples. While this skill-based strategy boosts ICL performance in larger models, its gains on small language models (SLMs) have been minimal, highlighting a performance gap in ICL capabilities. We investigate this gap and show that skill-based prompting can hurt SLM performance on easy questions by introducing unnecessary information, akin to cognitive overload. To address this, we introduce AdaptMI, an adaptive approach to selecting skill-based in-context Math Instructions for SLMs. Inspired by cognitive load theory from human pedagogy, our method only introduces skill-based examples when the model performs poorly. We further propose AdaptMI+, which adds examples targeted to the specific skills missing from the model's responses. On 5-shot evaluations across popular math benchmarks and five SLMs (1B--7B; Qwen, Llama), AdaptMI+ improves accuracy by up to 6% over naive skill-based strategies.

Real-world tasks often require large language models (LLMs) to learn from complex task-specific context rather than pretrained parametric knowledge. This capability remains a weakness of LLMs, while human annotation for such task contexts is expensive and difficult to scale. Public high-quality documents are an abundant alternative, but much of the public web has already been consumed during pretraining: training on such documents naively would reward memorization rather than context learning. In this work, we attempt to make use of high-quality public documents with small perturbations and empirically find that LLMs can successfully generate context-dependent reasoning traces and answers, which are then used to train a student model. Specifically, we construct a synthesis pipeline that (i) rewrites source documents to reduce memorization risk, (ii) generates questions and rubrics that require reasoning over the document, (iii) answers the questions with the document as context, and (iv) admits only samples that genuinely depend on the document. Without any human annotators, our pipeline generates about 10k samples from 3.5k documents, and the resulting student model substantially improves the performance on CL-bench. SFT raises a Qwen3.6-35B-A3B student from 13.7% to 22.8%, and a subsequent rubric-reward RL stage reaches 24.6%, on CL-bench comparable with a frontier model of over a trillion parameters, Qwen3.8-2.4T (23.9%). We also observe a broad transfer of improvements to long-context understanding, instruction following, and reasoning, while code generation and knowledge remain mostly flat. We hope this work provides a reproducible and scalable way to improve the ability of LLMs to learn from context, and to facilitate further research on context-grounded reasoning.

Large language models excel on objectively verifiable tasks such as math and programming, where evaluation reduces to unit tests or a single correct answer. In contrast, real-world enterprise work is often subjective and context-dependent: success hinges on organizational goals, user intent, and the quality of intermediate artifacts produced across long, multi-tool workflows. We introduce LH-Bench, a three-pillar evaluation design that moves beyond binary correctness to score autonomous, long-horizon execution on subjective enterprise tasks. The pillars are: (i) expert-grounded rubrics that give LLM judges the domain context needed to score subjective work, (ii) curated ground-truth artifacts that enable stepwise reward signals (e.g., chapter-level annotation for content tasks), and (iii) pairwise human preference evaluation for convergent validation. We show that domain-authored rubrics provide substantially more reliable evaluation signals than LLM-authored rubrics (kappa = 0.60 vs. 0.46), and that human preference judgments confirm the same top-tier separation (p < 0.05), evidence that expert-grounded evaluation can scale without sacrificing reliability. We release public datasets and report results on two environments: Figma-to-code (33 real .fig tasks against the Figma API via MCP) and Programmatic content (41 courses comprising 183 individually-evaluated chapters on a course platform serving 30+ daily users).

The Needle-in-a-Haystack (NIAH) benchmark is widely used to evaluate Large Language Models' (LLMs) ability to understand long contexts (LC). It evaluates the capability to identify query-relevant context within extensive query-irrelevant passages. Although this method serves as a widely accepted standard for evaluating long-context understanding, our findings suggest it may overestimate the true LC capability of LLMs. We demonstrate that even state-of-the-art models such as GPT-4o struggle to intactly incorporate given contexts made up of solely query-relevant ten sentences. In response, we introduce a novel benchmark, NeedleChain, where the context consists entirely of query-relevant information, requiring the LLM to fully grasp the input to answer correctly. Our benchmark allows for flexible context length and reasoning order, offering a more comprehensive analysis of LLM performance. Additionally, we propose an extremely simple yet compelling strategy to improve LC understanding capability of LLM: ROPE Contraction. Our experiments with various advanced LLMs reveal a notable disparity between their ability to process large contexts and their capacity to fully understand them. Source code and datasets are available at https://github.com/hyeonseokk/NeedleChain

Hallucination detection is increasingly important for evaluating and deploying large language models. However, existing methods are often compared across benchmarks that define and measure hallucination in substantially different ways, making reported scores difficult to interpret across settings. We study hallucination detection as a benchmark-conditioned measurement problem and organize existing benchmarks along three axes: provenance, which describes how model outputs and evidence conditions are obtained; typology, which captures the factuality and faithfulness failures represented by benchmark labels; and granularity, which specifies the unit at which hallucinations are annotated, predicted, and evaluated. Using this framework, we analyze how benchmark construction, annotation, and evaluation choices shape the underlying prediction task, and relate detector mechanisms and assumptions to the settings in which their performance is measured. We also examine span-level evaluation, where variation in annotation boundaries can substantially affect reported results, and propose Span Coverage F1 as a complementary containment-based metric. Our analysis clarifies what detector scores capture, highlights limitations of cross-benchmark comparisons, and motivates more explicit evaluation of generalization across benchmark conditions.

Long-horizon search agents must make multiple sequential actions (steps) to search, retrieve, verify, and integrate evidence to reach a final answer. However, existing methods for training these agents typically treat all steps within a trajectory uniformly during both supervised fine-tuning (SFT) and reinforcement learning (RL), failing to distinguish useful actions from erroneous or redundant ones. In this paper, we propose Answer-Backtracked Credit Assignment (ABC), a fine-grained credit assignment framework for training long-horizon search agents by converting sparse trajectory-level outcomes into dense step-level supervision that rewards useful actions (even in failed trajectories) while suppressing erroneous or redundant actions. Specifically, given a potentially obscure query and its corresponding ground-truth answer, ABC first performs Answer-Backtracked Clue Recovery, which traces back from the answer to recover intermediate clues required to solve the question. It then applies Clue-Anchored Step Scoring to evaluate each search step against these clues, converting sparse binary outcome supervision into dense step-level rewards. Based on these rewards, we develop ABC-SFT, which reweights the loss of each turn, and ABC-GRPO, which uses the step-level scores as rewards in GRPO. Building on this framework, we train ABSeeker based on Qwen3.5-4B with only 8.5k examples. ABSeeker achieves 37.3% on BrowseComp and 39.1% on BrowseComp-ZH. With context management, the scores further improve to 55.3% and 52.9%, respectively, significantly outperforming same-scale (4B) agents and even matching the performance of larger ones (approximately 30B). These results demonstrate the effectiveness of answer-backtracked step-level credit assignment for training long-horizon search agents.

We focus on the Source Free Object Detection (SFOD) problem, when source data is unavailable during adaptation, and the model must adapt to the unlabeled target domain. In medical imaging, several approaches have leveraged a semi-supervised student-teacher architecture to bridge domain discrepancy. Context imbalance in labeled training data and significant domain shifts between domains can lead to biased teacher models that produce inaccurate pseudolabels, degrading the student model's performance and causing a mode collapse. Class imbalance, particularly when one class significantly outnumbers another, leads to contextual bias. To tackle the problem of context bias and the significant performance drop of the student model in the SFOD setting, we introduce Grounded Teacher (GT) as a standard framework. In this study, we model contextual relationships using a dedicated relational context module and leverage it to mitigate inherent biases in the model. This approach enables us to apply augmentations to closely related classes, across and within domains, enhancing the performance of underrepresented classes while keeping the effect on dominant classes minimal. We further improve the quality of predictions by implementing an expert foundational branch to supervise the student model. We validate the effectiveness of our approach in mitigating context bias under the SFOD setting through experiments on three medical datasets supported by comprehensive ablation studies. All relevant resources, including preprocessed data, trained model weights, and code, are publicly available at this https://github.com/Tajamul21/Grounded_Teacher.

In long-context question answering (LCQA), determining the optimal amount of context for a given query is a significant challenge. Including too few passages may omit critical information, while including too many can introduce noise and reduce the quality of the answer. Traditional approaches, such as fixed Top-K retrieval and single-stage reranking, face the dilemma of selecting the right number of passages. This problem is particularly pronounced for factoid questions, which often require only a few specific pieces of evidence. To address this issue, we introduce Context-Picker, a reasoning-aware framework that shifts the paradigm from similarity-based ranking to minimal sufficient subset selection. Context-Picker treats context selection as a decision-making process optimized via a human-inspired, two-stage reinforcement learning schedule: a recall-oriented stage that prioritizes the coverage of reasoning chains, followed by a precision-oriented stage that aggressively prunes redundancy to distill a compact evidence set. To resolve reward sparsity, we propose an offline evidence distillation pipeline that mines "minimal sufficient sets" via a Leave-One-Out (LOO) procedure, providing dense, task-aligned supervision. Experiments on five long-context and multi-hop QA benchmarks demonstrate that Context-Picker significantly outperforms strong RAG baselines, achieving superior answer accuracy with comparable or reduced context lengths. Ablation studies indicate that the coarse-to-fine optimization schedule, the redundancy-aware reward shaping, and the rationale-guided format all contribute substantially to these gains.

Grounding language in the physical world requires AI systems to interpret references that emerge dynamically during conversation. While current vision-language models (VLMs) excel at static image tasks, they struggle to resolve ambiguous expressions in spontaneous, multi-turn dialogue. We address this gap by introducing (1) a benchmark for referential communication in dynamic 3D environments, built from 6.7 hours of egocentric VR interaction with synchronized speech, motion, gaze, and 3D scene geometry, and (2) a two-stage grounding pipeline that explicitly resolves conversational ambiguity before visual localization. The benchmark includes over 4,200 manually verified referring expressions spanning full, partitive, and pronominal types. Our contextual rewriting approach improves grounding performance by 11-22 percentage points on average, with a pure detector (GroundingDINO) reaching 56.7% on pronominals after rewriting, nearly double the best end-to-end baseline. Results demonstrate that decoupling linguistic reasoning from visual perception is more effective than end-to-end approaches for conversational grounding.

Many NLP tasks, such as summarization and extractive question answering, reduce to retrieving relevant content from documents under two constraints: coverage, retaining enough pertinent information to achieve some goal, and conciseness, removing as much irrelevant information as possible. Conformal prediction methods have been used to guarantee coverage, and must be optimized for conciseness through design of a score function. State-of-the-art scoring functions use hand-engineered LLM prompts asking the model to rate the importance of content, but manual prompt engineering is labor-intensive and task-specific. We introduce the Conformal Relevance framework which uses in-context learning example curation and ensembling to create a score function which maintains coverage while improving conciseness with minimal manual input. We demonstrate this framework's application on seven NLP tasks, and also theoretically study the impact of diversity for ensembled conformal scores, giving a complementarity condition that characterizes when ensembling improves worst-case sentence scores, and a saturation bound on ensemble improvement.

Context utilisation, the ability of Language Models (LMs) to incorporate relevant information from the provided context when generating responses, remains largely opaque to users, who cannot determine whether models draw from parametric memory or provided context, nor identify which specific context pieces inform the response. Highlight explanations (HEs) offer a natural solution as they can point the exact context pieces and tokens that influenced model outputs. However, no existing work evaluates their effectiveness in accurately explaining context utilisation. We address this gap by introducing the first gold standard HE evaluation framework for context attribution, using controlled test cases with known ground-truth context usage, which avoids the limitations of existing indirect proxy evaluations. To demonstrate the framework's broad applicability, we evaluate four HE methods -- three established techniques and MechLight, a mechanistic interpretability approach we adapt for this task -- across four context scenarios, four datasets, and five LMs. Overall, we find that MechLight performs best across all context scenarios. However, all methods struggle with longer contexts and exhibit positional biases, pointing to fundamental challenges in explanation accuracy that require new approaches to deliver reliable context utilisation explanations at scale.

AudioLLMs enable speech recognition conditioned on textual prompts such as domain descriptions or entity lists. However, it remains unclear whether these models genuinely utilise such context or rely on parametric knowledge learned during pretraining. Existing benchmarks cannot answer this question because they evaluate transcription under fixed prompting conditions and rarely include explicit contextual inputs. We introduce IndicContextEval, a 56-hour multilingual benchmark of natural speech from 555 speakers across 8 Indian languages and 23 professional domains. We design a 7-level prompting framework that progressively introduces contextual signals, including metadata, natural-language descriptions, entity lists in English and native script, and adversarial prompts with incorrect entities. Evaluating five models reveals substantial differences in context utilisation behaviour, highlighting the need for explicit evaluation of contextual grounding in AudioLLMs.

In-context learning (ICL), the ability of large language models to perform novel tasks by conditioning on a prompt with a few task examples, requires these examples to be informative about the test instance. The standard approach of independently ranking and selecting the most similar examples selects redundant examples while omitting important information. In this work, we show that BERTScore-Recall (BSR) selects better examples that demonstrate more of the salient aspects, e.g. reasoning patterns, of the test input. We further extend BSR and many standard metrics to easily optimizable set-level metrics, giving still better coverage of those salient aspects. On 15 datasets spanning 6 tasks and with 7 diverse LLMs, we show that (1) BSR is the superior metric for in-context example selection across the board, and (2) for compositional tasks, set selection using Set-BSR outperforms independent ranking by up to 17 points on average and, despite being training-free, surpasses methods that leverage task or LLM-specific training.

We conduct a systematic audit of three widely used reasoning benchmarks, SocialIQa, FauxPas-EAI, and ToMi, and uncover pervasive flaws in both benchmark items and evaluation methodology. Using five LLMs (GPT-{3, 3.5, 4, o1}, and LLaMA 3.1) as diagnostic tools, we identify structural, semantic, and pragmatic issues in benchmark design (e.g., duplicated items, ambiguous wording, and implausible answers), as well as scoring procedures that prioritize output form over reasoning process. Through systematic human annotation and re-evaluation on cleaned benchmark subsets, we find that model scores often improve not due to due to erratic surface wording variations and not to improved reasoning. Infact, further analyses show that model performance is highly sensitive to minor input variations such as context availability and phrasing, revealing that high scores may reflect alignment with format-specific cues rather than consistent inference based on the input. These findings challenge the validity of current benchmark-based claims about reasoning in LLMs, and highlight the need for evaluation protocols that assess reasoning as a process of drawing inference from available information, rather than as static output selection. We release audited data and evaluation tools to support more interpretable and diagnostic assessments of model reasoning.

While multimodal large language models (MLLMs) have demonstrated extraordinary vision-language understanding capabilities and shown potential to serve as general-purpose assistants, their abilities to solve instance-level visual-language problems beyond a single image warrant further exploration. In order to assess these unproven abilities of MLLMs, this paper proposes a new visual grounding task called multi-context visual grounding, which aims to localize instances of interest across multiple images based on open-ended text prompts. To facilitate this research, we meticulously construct a new dataset MC-Bench for benchmarking the visual grounding capabilities of MLLMs. MC-Bench features 2K high-quality and manually annotated samples, consisting of instance-level labeled image pairs and corresponding text prompts that indicate the target instances in the images. In total, there are three distinct styles of text prompts, covering 20 practical skills. We benchmark over 20 state-of-the-art MLLMs and foundation models with potential multi-context visual grounding capabilities. Our evaluation reveals a non-trivial performance gap between existing MLLMs and humans across all metrics. We also observe that existing MLLMs typically outperform foundation models without LLMs only on image-level metrics, and the specialist MLLMs trained on single images often struggle to generalize to multi-image scenarios. Moreover, a simple stepwise baseline integrating advanced MLLM and a detector can significantly surpass prior end-to-end MLLMs. We hope our MC-Bench and empirical findings can encourage the research community to further explore and enhance the untapped potentials of MLLMs in instance-level tasks, particularly in multi-image contexts. Project page: https://xuyunqiu.github.io/MC-Bench/.

Retrieval-Augmented Generation (RAG) enhances factual grounding in large language models (LLMs) by incorporating retrieved evidence, but LLM accuracy declines when long or noisy contexts exceed the model's effective attention span. Existing pre-generation filters rely on heuristics or uncalibrated LLM confidence scores, offering no statistical control over retained evidence. We evaluate and demonstrate context engineering through conformal prediction, a coverage-controlled filtering framework that removes irrelevant content while preserving recall of supporting evidence. Using both embedding- and LLM-based scoring functions, we test this approach on the NeuCLIR and RAGTIME collections. Conformal filtering consistently meets its target coverage, ensuring that a specified fraction of relevant snippets are retained, and reduces retained context by 2-3x relative to unfiltered retrieval. On NeuCLIR, downstream factual accuracy measured by ARGUE F1 improves under strict filtering and remains stable at moderate coverage, indicating that most discarded material is redundant or irrelevant. These results demonstrate that conformal prediction enables reliable, coverage-controlled context reduction in RAG, offering a model-agnostic and principled approach to context engineering.

We present GaRAGe, a large RAG benchmark with human-curated long-form answers and annotations of each grounding passage, allowing a fine-grained evaluation of whether LLMs can identify relevant grounding when generating RAG answers. Our benchmark contains 2366 questions of diverse complexity, dynamism, and topics, and includes over 35K annotated passages retrieved from both private document sets and the Web, to reflect real-world RAG use cases. This makes it an ideal test bed to evaluate an LLM's ability to identify only the relevant information necessary to compose a response, or provide a deflective response when there is insufficient information. Evaluations of multiple state-of-the-art LLMs on GaRAGe show that the models tend to over-summarise rather than (a) ground their answers strictly on the annotated relevant passages (reaching at most a Relevance-Aware Factuality Score of 60%), or (b) deflect when no relevant grounding is available (reaching at most 31% true positive rate in deflections). The F1 in attribution to relevant sources is at most 58.9%, and we show that performance is particularly reduced when answering time-sensitive questions and when having to draw knowledge from sparser private grounding sources.

We focus on grounding (i.e., localizing or linking) referring expressions in images, e.g., "largest elephant standing behind baby elephant". This is a general yet challenging vision-language task since it does not only require the localization of objects, but also the multimodal comprehension of context --- visual attributes (e.g., "largest", "baby") and relationships (e.g., "behind") that help to distinguish the referent from other objects, especially those of the same category. Due to the exponential complexity involved in modeling the context associated with multiple image regions, existing work oversimplifies this task to pairwise region modeling by multiple instance learning. In this paper, we propose a variational Bayesian method, called Variational Context, to solve the problem of complex context modeling in referring expression grounding. Our model exploits the reciprocal relation between the referent and context, i.e., either of them influences the estimation of the posterior distribution of the other, and thereby the search space of context can be greatly reduced, resulting in better localization of referent. We develop a novel cue-specific language-vision embedding network that learns this reciprocity model end-to-end. We also extend the model to the unsupervised setting where no annotation for the referent is available. Extensive experiments on various benchmarks show consistent improvement over state-of-the-art methods in both supervised and unsupervised settings.

Long-context capability is considered one of the most important abilities of LLMs, as a truly long context-capable LLM enables users to effortlessly process many originally exhausting tasks -- e.g., digesting a long-form document to find answers vs. directly asking an LLM about it. However, existing real-task-based long-context evaluation benchmarks have two major shortcomings. First, benchmarks like LongBench often do not provide proper metrics to separate long-context performance from the model's baseline ability, making cross-model comparison unclear. Second, such benchmarks are usually constructed with fixed input lengths, which limits their applicability across different models and fails to reveal when a model begins to break down. To address these issues, we introduce a length-controllable long-context benchmark and a novel metric that disentangles baseline knowledge from true long-context capabilities. Experiments demonstrate the superiority of our approach in effectively evaluating LLMs.

Language grounding to vision is an active field of research aiming to enrich text-based representations of word meanings by leveraging perceptual knowledge from vision. Despite many attempts at language grounding, it is still unclear how to effectively inject visual knowledge into the word embeddings of a language in such a way that a proper balance of textual and visual knowledge is maintained. Some common concerns are the following. Is visual grounding beneficial for abstract words or is its contribution only limited to concrete words? What is the optimal way of bridging the gap between text and vision? How much do we gain by visually grounding textual embeddings? The present study addresses these questions by proposing a simple yet very effective grounding approach for pre-trained word embeddings. Our model aligns textual embeddings with vision while largely preserving the distributional statistics that characterize word use in text corpora. By applying a learned alignment, we are able to generate visually grounded embeddings for unseen words, including abstract words. A series of evaluations on word similarity benchmarks shows that visual grounding is beneficial not only for concrete words, but also for abstract words. We also show that our method for visual grounding offers advantages for contextualized embeddings, but only when these are trained on corpora of relatively modest size. Code and grounded embeddings for English are available at https://github.com/Hazel1994/Visually_Grounded_Word_Embeddings_2.

The long-context capabilities of large language models (LLMs) have been a hot topic in recent years. To evaluate the performance of LLMs in different scenarios, various assessment benchmarks have emerged. However, as most of these benchmarks focus on identifying key information to answer questions, which mainly requires the retrieval ability of LLMs, these benchmarks can partially represent the reasoning performance of LLMs from large amounts of information. Meanwhile, although LLMs often claim to have context windows of 32k, 128k, 200k, or even longer, these benchmarks fail to reveal the actual supported length of these LLMs. To address these issues, we propose the LongIns benchmark dataset, a challenging long-context instruction-based exam for LLMs, which is built based on the existing instruction datasets. Specifically, in our LongIns, we introduce three evaluation settings: Global Instruction & Single Task (GIST), Local Instruction & Single Task (LIST), and Local Instruction & Multiple Tasks (LIMT). Based on LongIns, we perform comprehensive evaluations on existing LLMs and have the following important findings: (1). The top-performing GPT-4 with 128k context length performs poorly on the evaluation context window of 16k in our LongIns. (2). For the multi-hop reasoning ability of many existing LLMs, significant efforts are still needed under short context windows (less than 4k).

Multimodal large language models can emit localized predictions, bounding boxes for objects and temporal windows for video and audio events, but they hallucinate these regions prolifically. The model's own token log-probabilities are nearly uninformative: they conflate grounding quality with input ambiguity, and coordinate tokens become near-deterministic once the model commits. We propose Multi-Token Localized Attention (MTLA): a training-free, post-hoc score that measures how strongly a prediction's tokens attend to the region they claim. Prior attention-based detectors, which sum attention over the entire input modality and read a single response token, are weaker special cases; we show that summing only within the claimed region and aggregating across all prediction tokens recovers a stronger grounding signal. The same recipe applies almost trivially to other modalities and tasks: object detection in images and temporal localization in video and audio. Across multiple MLLM families and three modalities, MTLA improves hallucination AUROC by +7 to +38 over the best prior training-free baseline. Used as a confidence score for re-ranking, it nearly doubles the zero-shot COCO detection AP of an open-source 8B generalist (from 20.4 to 37.0), narrowing the gap to supervised detectors without any task-specific training.

Multimodal large language models (MLLMs) are increasingly used for real-world tasks involving multi-step reasoning and long-form generation, where reliability requires grounding model outputs in heterogeneous input sources and verifying individual factual claims. However, existing multimodal grounding benchmarks and evaluation methods focus on simplified, observation-based scenarios or limited modalities and fail to assess attribution in complex multimodal reasoning. We introduce MuRGAt (Multimodal Reasoning with Grounded Attribution), a benchmark for evaluating fact-level multimodal attribution in settings that require reasoning beyond direct observation. Given inputs spanning video, audio, and other modalities, MuRGAt requires models to generate answers with explicit reasoning and precise citations, where each citation specifies both modality and temporal segments. To enable reliable assessment, we introduce an automatic evaluation framework that strongly correlates with human judgments. Benchmarking with human and automated scores reveals that even strong MLLMs frequently hallucinate citations despite correct reasoning. Moreover, we observe a key trade-off: increasing reasoning depth or enforcing structured grounding often degrades accuracy, highlighting a significant gap between internal reasoning and verifiable attribution.

Contextual ASR can benefit from semantic cues or from target words explicitly provided in the context. We introduce HearInContext, a Mandarin--English benchmark that pairs shared synthetic speech with assistant replies supporting different interpretations. The benchmark comprises 3,764 semantic test cases built around homophones. Implicit contexts exclude candidate words; explicit contexts name the target. No-context and unrelated-context controls measure the benefit of relevant history and sensitivity to irrelevant history. Context-capable models benefit from implicit cues but achieve higher target recall with explicit hints. Fine-tuning Qwen3-ASR-1.7B improves implicit-context target recall by 11.0 and 11.5 percentage points in Mandarin and English, respectively, while absolute CER/WER changes on AISHELL-1 and LibriSpeech remain below 0.1 percentage points. Gains extend to explicit conditions excluded from fine-tuning and to Mandarin hotword recognition on real recordings.

Language model users often issue queries that lack specification, where the context under which a query was issued -- such as the user's identity, the query's intent, and the criteria for a response to be useful -- is not explicit. For instance, a good response to a subjective query like "What book should I read next?" would depend on the user's preferences, and a good response to an open-ended query like "How do antibiotics work against bacteria?" would depend on the user's expertise. This makes evaluation of responses to such queries an ill-posed task, as evaluators may make arbitrary judgments about the response quality. To remedy this, we present contextualized evaluations, a protocol that synthetically constructs context surrounding an underspecified query and provides it during evaluation. We find that the presence of context can 1) alter conclusions drawn from evaluation, even flipping win rates between model pairs, 2) nudge evaluators to make fewer judgments based on surface-level criteria, like style, and 3) provide new insights about model behavior across diverse contexts. Specifically, our procedure uncovers an implicit bias towards WEIRD contexts in models' "default" responses and we find that models are not equally sensitive to following different contexts, even when they are provided in prompts.

Graphical user interface (GUI) grounding, the ability to map natural language instructions to specific actions on graphical user interfaces, remains a critical bottleneck in computer use agent development. Current benchmarks oversimplify grounding tasks as short referring expressions, failing to capture the complexity of real-world interactions that require software commonsense, layout understanding, and fine-grained manipulation capabilities. To address these limitations, we introduce OSWorld-G, a comprehensive benchmark comprising 564 finely annotated samples across diverse task types including text matching, element recognition, layout understanding, and precise manipulation. Additionally, we synthesize and release the largest computer use grounding dataset Jedi, which contains 4 million examples through multi-perspective decoupling of tasks. Our multi-scale models trained on Jedi demonstrate its effectiveness by outperforming existing approaches on ScreenSpot-v2, ScreenSpot-Pro, and our OSWorld-G. Furthermore, we demonstrate that improved grounding with Jedi directly enhances agentic capabilities of general foundation models on complex computer tasks, improving from 5% to 27% on OSWorld. Through detailed ablation studies, we identify key factors contributing to grounding performance and verify that combining specialized data for different interface elements enables compositional generalization to novel interfaces. All benchmark, data, checkpoints, and code are open-sourced and available at https://osworld-grounding.github.io.

Large language models (LLMs) have had a huge impact on society due to their impressive capabilities and vast knowledge of the world. Various applications and tools have been created that allow users to interact with these models in a black-box scenario. However, one limitation of this scenario is that users cannot modify the internal knowledge of the model, and the only way to add or modify internal knowledge is by explicitly mentioning it to the model during the current interaction. This learning process is called in-context training, and it refers to training that is confined to the user's current session or context. In-context learning has significant applications, but also has limitations that are seldom studied. In this paper, we present a study that shows how the model can suffer from interference between information that continually flows in the context, causing it to forget previously learned knowledge, which can reduce the model's performance. Along with showing the problem, we propose an evaluation benchmark based on the bAbI dataset.

Recent video multimodal large language models achieve impressive results across various benchmarks. However, current evaluations suffer from two critical limitations: (1) inflated scores can mask deficiencies in fine-grained visual understanding and reasoning, and (2) answer correctness is often measured without verifying whether models identify the precise spatio-temporal evidence supporting their predictions. To address this, we present VideoZeroBench, a hierarchical benchmark designed for challenging long-video question answering that rigorously verifies spatio-temporal evidence. It comprises 500 manually annotated questions across 13 domains, paired with temporal intervals and spatial bounding boxes as evidence. To disentangle answering generation, temporal grounding, and spatial grounding, we introduce a five-level evaluation protocol that progressively tightens evidence requirements. Experiments show that even Gemini-3-Pro correctly answers fewer than 17% of questions under the standard end-to-end QA setting (Level-3). When grounding constraints are imposed, performance drops sharply: No model exceeds 1% accuracy when both correct answering and accurate spatio-temporal localization are required (Level-5), with most failing to achieve any correct grounded predictions. These results expose a significant gap between surface-level answer correctness and genuine evidence-based reasoning, revealing that grounded video understanding remains a bottleneck for long-video QA. We further analyze performance across minimal evidence spans, atomic abilities, and inference paradigms, providing insights for future research in grounded video reasoning. The benchmark and code will be made publicly available.

Large Language Models (LLMs) have the impressive ability to perform in-context learning (ICL) from only a few examples, but the success of ICL varies widely from task to task. Thus, it is important to quickly determine whether ICL is applicable to a new task, but directly evaluating ICL accuracy can be expensive in situations where test data is expensive to annotate -- the exact situations where ICL is most appealing. In this paper, we propose the task of ICL accuracy estimation, in which we predict the accuracy of an LLM when doing in-context learning on a new task given only unlabeled test data for that task. To perform ICL accuracy estimation, we propose a method that trains a meta-model using LLM confidence scores as features. We compare our method to several strong accuracy estimation baselines on a new benchmark that covers 4 LLMs and 3 task collections. The meta-model improves over all baselines across 8 out of 12 settings and achieves the same estimation performance as directly evaluating on 40 collected labeled test examples per task. At the same time, no existing approach provides an accurate and reliable ICL accuracy estimation in every setting, highlighting the need for better ways to measure the uncertainty of LLM predictions.

Video interactive storytelling enables viewers to actively steer how a video unfolds. However, once we allow viewers to intervene during generation, a new challenge arises: The viewer's request can have latent dependencies on both the grounding source and the current rendered video state. These dependencies may not be explicitly stated in any individual input, but emerge only when the source, rendered history, and new viewer intent are considered jointly. Existing interactive video generation systems primarily emphasize following viewer instructions, while source-grounded video generation methods focus on aligning generated content with an external narrative or knowledge source. This leaves a fundamental question underexplored: What context should a generation model ground on during interactive continuation, and how can heterogeneous, unstructured inputs be transformed into such grounding context? In this work, we formulate contextual grounding as the process of transforming heterogeneous input context into an executable constraint model for video generation. To address this challenge, we introduce VIS-Ground, which performs Structured Context Abstraction to recover grounded states and cross-context dependencies, Generation Constraints Induction to project relevant dependencies into candidate-specific constraints, and Constrained Video Generation to enforce these constraints through planning, verification, revision, and rendering. Across three video generation backbones, VIS-Ground consistently achieves the highest overall composite score, reaching an average absolute improvement of 10.3 points over the strongest per-backbone baselines. Detailed analysis further shows gains across both narrative and knowledge grounding, and reveals remaining challenges in dependency extraction, and faithful realization during video rendering.

Benchmark datasets are central to evaluating Large Language Models (LLMs), yet they are typically conceived as monolithic tasks, obscuring substantial variation in the demands of individual samples. We introduce a dataset-centric meta-evaluation framework that audits benchmark datasets at the sample level along five latent dimensions: 1. Cognitive and Knowledge Demands, 2. Language and Content Quality, 3. Task Properties, 4. Context, and 5. Ethics, Safety, and Fairness. Applying this framework, we annotate five influential benchmarks -- MMLU, ARC, WinoGrande, HellaSwag, and TruthfulQA -- revealing pronounced internal heterogeneity that is not captured by aggregate accuracy scores. We show how these annotations enable criterion-driven orchestration of composite benchmark subsets across datasets, supporting targeted evaluation of model capabilities such as Reasoning Depth or Ethical Sensitivity. This approach reframes benchmark evaluation as dataset introspection, providing a principled methodology for analyzing and re-composing existing benchmarks to better reflect diverse evaluation needs.

Despite the impressive performance on information-seeking tasks, large language models (LLMs) still struggle with hallucinations. Attributed LLMs, which augment generated text with in-line citations, have shown potential in mitigating hallucinations and improving verifiability. However, current approaches suffer from suboptimal citation quality due to their reliance on in-context learning. Furthermore, the practice of citing only coarse document identifiers makes it challenging for users to perform fine-grained verification. In this work, we introduce FRONT, a training framework designed to teach LLMs to generate Fine-Grained Grounded Citations. By grounding model outputs in fine-grained supporting quotes, these quotes guide the generation of grounded and consistent responses, not only improving citation quality but also facilitating fine-grained verification. Experiments on the ALCE benchmark demonstrate the efficacy of FRONT in generating superior grounded responses and highly supportive citations. With LLaMA-2-7B, the framework significantly outperforms all the baselines, achieving an average of 14.21% improvement in citation quality across all datasets, even surpassing ChatGPT.

Large language models (LLMs), renowned for their powerful conversational abilities, are widely recognized as exceptional tools in the field of education, particularly in the context of automated intelligent instruction systems for language learning. In this paper, we propose a scoring system based on LLMs, motivated by their positive impact on text-related scoring tasks. Specifically, the speech encoder first maps the learner's speech into contextual features. The adapter layer then transforms these features to align with the text embedding in latent space. The assessment task-specific prefix and prompt text are embedded and concatenated with the features generated by the modality adapter layer, enabling the LLMs to predict accuracy and fluency scores. Our experiments demonstrate that the proposed scoring systems achieve competitive results compared to the baselines on the Speechocean762 datasets. Moreover, we also conducted an ablation study to better understand the contributions of the prompt text and training strategy in the proposed scoring system.

While in-context Learning (ICL) has proven to be an effective technique to improve the performance of Large Language Models (LLMs) in a variety of complex tasks, notably in translating natural language questions into Structured Query Language (NL2SQL), the question of how to select the most beneficial demonstration examples remains an open research problem. While prior works often adapted off-the-shelf encoders to retrieve examples dynamically, an inherent discrepancy exists in the representational capacities between the external retrievers and the LLMs. Further, optimizing the selection of examples is a non-trivial task, since there are no straightforward methods to assess the relative benefits of examples without performing pairwise inference. To address these shortcomings, we propose DeTriever, a novel demonstration retrieval framework that learns a weighted combination of LLM hidden states, where rich semantic information is encoded. To train the model, we propose a proxy score that estimates the relative benefits of examples based on the similarities between output queries. Experiments on two popular NL2SQL benchmarks demonstrate that our method significantly outperforms the state-of-the-art baselines on one-shot NL2SQL tasks.

Hallucination detection for retrieval-augmented generation (RAG) is usually evaluated on natural-language document evidence. However, grounded generation systems increasingly rely on structured inputs: source code, developer-tool output, markdown documents, tables, and repository metadata. We introduce a unified benchmark for span-level hallucination detection over code, tool output, structured documents, and existing natural-language RAG datasets. The benchmark is built by starting from grounded correct answers, injecting localized hallucinations with exact character labels, and validating the code test split with evidence-based review. Our fine-tuned Qwen3.5-2B detector reaches 0.689 span-F1 on the unified test set and 0.60 on the code-agent source, where it substantially outperforms LettuceDetect-large (0.17) and the strongest zero-shot LLM judges we evaluated (at most 0.22). The same model remains competitive on established natural-language benchmarks, with 81.8 RAGTruth example-F1 and 0.724 English PsiloQA IoU.

Vision-language models can perform new tasks without parameter updates through in-context learning (ICL), whose core mechanism is utilizing the support set for task induction. In the standard ICL setting, once the task is induced, its decision criterion remains fixed. However, in real-world applications, many tasks exhibit a stable high-level intent, while their decision criteria shift according to specific requirements. Thus, we introduce a new setting, denoted as Criterion-Conditional In-Context Learning (CC-ICL), where models must infer the latent criterion from context and adjust predictions accordingly under fixed task semantics. To evaluate this capability, we propose two complementary metrics, Criterion Invariance and Criterion Sensitivity, capturing the model's robustness and adaptability under criterion shifts. We further construct CC-Bench, a multi-domain benchmark that supports evaluation under the CC-ICL setting. By employing a dual-level data hierarchy, CC-Bench enables legitimate ground-truth variation conditioned on the active criterion even when the task remains fixed. Experiments on CC-Bench reveal that most models exhibit a rigid boundary bias, struggling to align their decisions with the latent criterion. We also find that even a simple multi-criterion training strategy can significantly reduce this bias, improving Criterion Sensitivity and enabling 7B-scale models to surpass proprietary models without degrading general multimodal performance.

We measure the performance of in-context learning as a function of task novelty and difficulty for open and closed questions. For that purpose, we created a novel benchmark consisting of hard scientific questions, each paired with a context of various relevancy. We show that counter-intuitively, a context that is more aligned with the topic does not always help more than a less relevant context. This effect is especially visible for open questions and questions of high difficulty or novelty. This result reveals a fundamental difference between the treatment of close-form and open-form questions by large-language models and shows a need for a more robust evaluation of in-context learning on the variety of different types of questions. It also poses a new question of how to optimally select a context for large language models, especially in the context of Retrieval Augmented Generation (RAG) systems. Our results suggest that the answer to this question can be highly application-dependent and might be contingent on factors including the format of the question, the perceived difficulty level of the questions, and the novelty or popularity of the information we seek.

AI agents can now carry out data-driven scientific analyses end to end, and benchmarks assess them by giving an agent a question and a dataset and scoring its final answer against a fixed key. These benchmarks assume that a correct answer was derived from the supplied data, a property we call evidence grounding. However, an agent can also reach the key from prior knowledge or by ruling out the other options, and a score based on a single run cannot tell these cases apart. We show how to test this assumption and find that it often fails. For each question, we build versions of its data files in which the evidence for the answer is left intact, withdrawn or reversed, check each edit with a pre-registered reference statistic, and run the same agent on every version. We then measure evidence-grounded accuracy, which credits a correct answer only if the agent also responds when the evidence is withdrawn and follows it when it is reversed. We evaluate three agent scaffolds and five models on 18 single-cell questions from BAISBench and four synthetic problems from GeneBench-Pro. On the single-cell questions, the Claude agents are 95% accurate and answer 83% correctly without any data, but their evidence-grounded accuracy is only 41%. Hiding gene names raises the share of runs that follow reversed evidence from 58% to 93% on ten gene tasks, suggesting that prior knowledge competes with the supplied data. The benchmark score and LLM judges can also reward answers that ignore the changed evidence. Measuring scientific intelligence rather than recall therefore requires checking whether answers follow the evidence and whether scores reward them for it.

Progress on training and architecture strategies has enabled LLMs with millions of tokens in context length. However, empirical evidence suggests that such long-context LLMs can consume far more text than they can reliably use. On the other hand, it has been shown that inference-time compute can be used to scale performance of LLMs, often by generating thinking tokens, on challenging tasks involving multi-step reasoning. Through controlled experiments on sandbox long-context tasks, we find that such inference-time strategies show rapidly diminishing returns and fail at long context. We attribute these failures to score dilution, a phenomenon inherent to static self-attention. Further, we show that current inference-time strategies cannot retrieve relevant long-context signals under certain conditions. We propose a simple method that, through targeted gradient updates on the given context, provably overcomes limitations of static self-attention. We find that this shift in how inference-time compute is spent leads to consistently large performance improvements across models and long-context benchmarks. Our method leads to large 12.6 and 14.1 percentage point improvements for Qwen3-4B on average across subsets of LongBench-v2 and ZeroScrolls benchmarks. The takeaway is practical: for long context, a small amount of context-specific training is a better use of inference compute than current inference-time scaling strategies like producing more thinking tokens.

GUI grounding, the task of identifying a precise location on an interface image from a natural language query, plays a crucial role in enhancing the capabilities of Vision-Language Model (VLM) agents. While general VLMs, such as GPT-4V, demonstrate strong performance across various tasks, their proficiency in GUI grounding remains suboptimal. Recent studies have focused on fine-tuning these models specifically for one-shot GUI grounding, yielding significant improvements over baseline performance. We introduce a visual prompting framework called Iterative Narrowing (IN) to further enhance the performance of both general and fine-tuned models in GUI grounding. For evaluation, we tested our method on a comprehensive benchmark comprising different UI platforms.

Recent advancements in AI have led to the development of large multimodal models (LMMs) capable of processing complex tasks involving joint reasoning over text and visual content in the image (e.g., navigating maps in public places). This paper introduces ConTextual, a novel benchmark comprising instructions designed explicitly to evaluate LMMs' ability to perform context-sensitive text-rich visual reasoning. ConTextual emphasizes diverse real-world scenarios (e.g., time-reading, navigation, shopping and more) demanding a deeper understanding of the interactions between textual and visual elements. Our findings reveal a significant performance gap of 30.8% between the best-performing LMM, GPT-4V(ision), and human capabilities using human evaluation indicating substantial room for improvement in context-sensitive text-rich visual reasoning. Notably, while GPT-4V excelled in abstract categories like meme and quote interpretation, its overall performance still lagged behind humans. In addition to human evaluations, we also employed automatic evaluation metrics using GPT-4, uncovering similar trends in performance disparities. We also perform a fine-grained evaluation across diverse visual contexts and provide qualitative analysis which provides a robust framework for future advancements in the LMM design. https://con-textual.github.io/

Current language models (LMs) excel at reasoning over prompts using pre-trained knowledge. However, real-world tasks are far more complex and context-dependent: models must learn from task-specific context and leverage new knowledge beyond what is learned during pre-training to reason and resolve tasks. We term this capability context learning, a crucial ability that humans naturally possess but has been largely overlooked. To this end, we introduce CL-bench, a real-world benchmark consisting of 500 complex contexts, 1,899 tasks, and 31,607 verification rubrics, all crafted by experienced domain experts. Each task is designed such that the new content required to resolve it is contained within the corresponding context. Resolving tasks in CL-bench requires models to learn from the context, ranging from new domain-specific knowledge, rule systems, and complex procedures to laws derived from empirical data, all of which are absent from pre-training. This goes far beyond long-context tasks that primarily test retrieval or reading comprehension, and in-context learning tasks, where models learn simple task patterns via instructions and demonstrations. Our evaluations of ten frontier LMs find that models solve only 17.2% of tasks on average. Even the best-performing model, GPT-5.1, solves only 23.7%, revealing that LMs have yet to achieve effective context learning, which poses a critical bottleneck for tackling real-world, complex context-dependent tasks. CL-bench represents a step towards building LMs with this fundamental capability, making them more intelligent and advancing their deployment in real-world scenarios.

Processing and reasoning over long contexts is crucial for many practical applications of Large Language Models (LLMs), such as document comprehension and agent construction. Despite recent strides in making LLMs process contexts with more than 100K tokens, there is currently a lack of a standardized benchmark to evaluate this long-context capability. Existing public benchmarks typically focus on contexts around 10K tokens, limiting the assessment and comparison of LLMs in processing longer contexts. In this paper, we propose inftyBench, the first LLM benchmark featuring an average data length surpassing 100K tokens. inftyBench comprises synthetic and realistic tasks spanning diverse domains, presented in both English and Chinese. The tasks in inftyBench are designed to require well understanding of long dependencies in contexts, and make simply retrieving a limited number of passages from contexts not sufficient for these tasks. In our experiments, based on inftyBench, we evaluate the state-of-the-art proprietary and open-source LLMs tailored for processing long contexts. The results indicate that existing long context LLMs still require significant advancements to effectively process 100K+ context. We further present three intriguing analyses regarding the behavior of LLMs processing long context.

Visual grounding is a ubiquitous building block in many vision-language tasks and yet remains challenging due to large variations in visual and linguistic features of grounding entities, strong context effect and the resulting semantic ambiguities. Prior works typically focus on learning representations of individual phrases with limited context information. To address their limitations, this paper proposes a language-guided graph representation to capture the global context of grounding entities and their relations, and develop a cross-modal graph matching strategy for the multiple-phrase visual grounding task. In particular, we introduce a modular graph neural network to compute context-aware representations of phrases and object proposals respectively via message propagation, followed by a graph-based matching module to generate globally consistent localization of grounding phrases. We train the entire graph neural network jointly in a two-stage strategy and evaluate it on the Flickr30K Entities benchmark. Extensive experiments show that our method outperforms the prior state of the arts by a sizable margin, evidencing the efficacy of our grounding framework. Code is available at "https://github.com/youngfly11/LCMCG-PyTorch".

Code review is a cornerstone of software quality assurance, and recent advances in Large Language Models (LLMs) have shown promise in its automation. However, existing benchmarks for LLM-based code review face three major limitations. Lack of semantic context: most benchmarks provide only code diffs without textual information such as issue descriptions, which are crucial for understanding developer intent. Data quality issues: without rigorous validation, many samples are noisy-e.g., reviews on outdated or irrelevant code-reducing evaluation reliability. Coarse granularity: most benchmarks operate at the file or commit level, overlooking the fine-grained, line-level reasoning essential for precise review. We introduce ContextCRBench, a high-quality, context-rich benchmark for fine-grained LLM evaluation in code review. Our construction pipeline comprises: Raw Data Crawling, collecting 153.7K issues and pull requests from top-tier repositories; Comprehensive Context Extraction, linking issue-PR pairs for textual context and extracting the full surrounding function or class for code context; and Multi-stage Data Filtering, combining rule-based and LLM-based validation to remove outdated, malformed, or low-value samples, resulting in 67,910 context-enriched entries. ContextCRBench supports three evaluation scenarios aligned with the review workflow: hunk-level quality assessment, line-level defect localization, and line-level comment generation. Evaluating eight leading LLMs (four closed-source and four open-source) reveals that textual context yields greater performance gains than code context alone, while current LLMs remain far from human-level review ability. Deployed at ByteDance, ContextCRBench drives a self-evolving code review system, improving performance by 61.98% and demonstrating its robustness and industrial utility. https://github.com/kinesiatricssxilm14/ContextCRBench.

In-context learning (ICL) i.e. showing LLMs only a few task-specific demonstrations has led to downstream gains with no task-specific fine-tuning required. However, LLMs are sensitive to the choice of prompts, and therefore a crucial research question is how to select good demonstrations for ICL. One effective strategy is leveraging semantic similarity between the ICL demonstrations and test inputs by using a text retriever, which however is sub-optimal as that does not consider the LLM's existing knowledge about that task. From prior work (Min et al., 2022), we already know that labels paired with the demonstrations bias the model predictions. This leads us to our hypothesis whether considering LLM's existing knowledge about the task, especially with respect to the output label space can help in a better demonstration selection strategy. Through extensive experimentation on three text classification tasks, we find that it is beneficial to not only choose semantically similar ICL demonstrations but also to choose those demonstrations that help resolve the inherent label ambiguity surrounding the test example. Interestingly, we find that including demonstrations that the LLM previously mis-classified and also fall on the test example's decision boundary, brings the most performance gain.

Question answering (QA) has become an important application in the advanced development of large language models. General pre-trained large language models for question-answering are not trained to properly understand the knowledge or terminology for a specific domain, such as finance, healthcare, education, and customer service for a product. To better cater to domain-specific understanding, we build an in-house question-answering system for Adobe products. We propose a novel framework to compile a large question-answer database and develop the approach for retrieval-aware finetuning of a Large Language model. We showcase that fine-tuning the retriever leads to major improvements in the final generation. Our overall approach reduces hallucinations during generation while keeping in context the latest retrieval information for contextual grounding.

Multimodal language models can exhibit hallucinations in their outputs, which limits their reliability. The ability to automatically detect these errors is important for mitigating them, but has been less explored and existing efforts do not localize hallucinations, instead framing this as a classification task. In this work, we first pose multimodal hallucination detection as a sequence labeling task where models must localize hallucinated text spans and present a strong baseline model. Given the high cost of human annotations for this task, we propose an approach to improve the sample efficiency of these models by creating corrupted grounding data, which we use for pre-training. Leveraging phrase grounding data, we generate hallucinations to replace grounded spans and create hallucinated text. Experiments show that pre-training on this data improves sample efficiency when fine-tuning, and that the learning signal from the grounding data plays an important role in these improvements.

Question Answering (QA) accounts for a significant portion of LLM usage "in the wild". However, LLMs sometimes produce false or misleading responses, also known as "hallucinations". Therefore, grounding the generated answers in contextually provided information -- i.e., providing evidence for the generated text -- is paramount for LLMs' trustworthiness. Providing this information is the task of context attribution. In this paper, we systematically study LLM-based approaches for this task, namely we investigate (i) zero-shot inference, (ii) LLM ensembling, and (iii) fine-tuning of small LMs on synthetic data generated by larger LLMs. Our key contribution is SynQA: a novel generative strategy for synthesizing context attribution data. Given selected context sentences, an LLM generates QA pairs that are supported by these sentences. This leverages LLMs' natural strengths in text generation while ensuring clear attribution paths in the synthetic training data. We show that the attribution data synthesized via SynQA is highly effective for fine-tuning small LMs for context attribution in different QA tasks and domains. Finally, with a user study, we validate the usefulness of small LMs (fine-tuned on synthetic data from SynQA) in context attribution for QA.

Given a textual phrase and an image, the visual grounding problem is the task of locating the content of the image referenced by the sentence. It is a challenging task that has several real-world applications in human-computer interaction, image-text reference resolution, and video-text reference resolution. In the last years, several works have addressed this problem by proposing more and more large and complex models that try to capture visual-textual dependencies better than before. These models are typically constituted by two main components that focus on how to learn useful multi-modal features for grounding and how to improve the predicted bounding box of the visual mention, respectively. Finding the right learning balance between these two sub-tasks is not easy, and the current models are not necessarily optimal with respect to this issue. In this work, we propose a loss function based on bounding boxes classes probabilities that: (i) improves the bounding boxes selection; (ii) improves the bounding boxes coordinates prediction. Our model, although using a simple multi-modal feature fusion component, is able to achieve a higher accuracy than state-of-the-art models on two widely adopted datasets, reaching a better learning balance between the two sub-tasks mentioned above.

Large Language Models (LLMs) have inherent limitations of faithfulness and factuality, commonly referred to as hallucinations. Several benchmarks have been developed that provide a test bed for factuality evaluation within the context of English-centric datasets, while relying on supplementary informative context like web links or text passages but ignoring the available structured factual resources. To this end, Knowledge Graphs (KGs) have been identified as a useful aid for hallucination mitigation, as they provide a structured way to represent the facts about entities and their relations with minimal linguistic overhead. We bridge the lack of KG paths and multilinguality for factual language modeling within the existing hallucination evaluation benchmarks and propose a KG-based multilingual, multihop benchmark called MultiHal framed for generative text evaluation. As part of our data collection pipeline, we mined 140k KG-paths from open-domain KGs, from which we pruned noisy KG-paths, curating a high-quality subset of 25.9k. Our baseline evaluation shows an absolute scale increase by approximately 0.12 to 0.36 points for the semantic similarity score in KG-RAG over vanilla QA across multiple languages and multiple models, demonstrating the potential of KG integration. We anticipate MultiHal will foster future research towards several graph-based hallucination mitigation and fact-checking tasks.

This paper introduces a comprehensive system for detecting hallucinations in large language model (LLM) outputs in enterprise settings. We present a novel taxonomy of LLM responses specific to hallucination in enterprise applications, categorizing them into context-based, common knowledge, enterprise-specific, and innocuous statements. Our hallucination detection model HDM-2 validates LLM responses with respect to both context and generally known facts (common knowledge). It provides both hallucination scores and word-level annotations, enabling precise identification of problematic content. To evaluate it on context-based and common-knowledge hallucinations, we introduce a new dataset HDMBench. Experimental results demonstrate that HDM-2 out-performs existing approaches across RagTruth, TruthfulQA, and HDMBench datasets. This work addresses the specific challenges of enterprise deployment, including computational efficiency, domain specialization, and fine-grained error identification. Our evaluation dataset, model weights, and inference code are publicly available.

Long contexts challenge transformers: attention scores dilute across thousands of tokens, critical information is often lost in the middle, and models struggle to adapt to novel patterns at inference time. Recent work on test-time adaptation addresses this by maintaining a form of working memory -- transient parameters updated on the current context -- but existing approaches rely on uniform write policies that waste computation on low-utility regions and suffer from high gradient variance across semantically heterogeneous contexts. In this work, we reframe test-time adaptation as a budget-constrained memory consolidation problem, focusing on which parts of the context should be consolidated into working memory under limited computation. We propose Gdwm (Gated Differentiable Working Memory), a framework that introduces a write controller to gate the consolidation process. The controller estimates Contextual Utility, an information-theoretic measure of long-range contextual dependence, and allocates gradient steps accordingly while maintaining global coverage. Experiments on ZeroSCROLLS and LongBench v2 demonstrate that Gdwm achieves comparable or superior performance with 4times fewer gradient steps than uniform baselines, establishing a new efficiency-performance Pareto frontier for test-time adaptation.