Large Vision-Language Models (LVLMs) often suffer from object hallucination, generating objects that are absent from the image. Prior work largely attributes this to insufficient visual attention. However, we find that both real and hallucinated objects receive equally strong visual attention in the model's mid-to-late layers, suggesting that the key issue may not be how much the model attends, but what it attends to and why. To this end, we decode the visual features of high-attention regions using Logit Lens, and observe that regions corresponding to real objects can be correctly decoded to the target object tokens, whereas those for hallucinated objects cannot. Building on this, we identify two hallucination mechanisms: (i) visual uncertainty, triggered by semantically similar or confusable regions; masking these regions eliminates the hallucination. (ii) contextual prior, triggered by strong co-occurrence priors; even when the initially attended region is masked, the hallucination persists and attention drifts to other regions. Based on these findings, we propose a simple yet effective training-free Detect-Mitigate framework comprising a Logit-Lens Consistency Check to detect hallucination and targeted remedies: High-Attention Regions Masking (HARM) for visual uncertainty hallucination, and Visual Evidence Enhanced Decoding (VEED) for contextual prior hallucination. Our approach achieves state-of-the-art results on multiple hallucination benchmarks. Code will be available.
Current Multimodal Large Language Models (MLLMs) can process diverse sensory inputs, yet their reasoning remains heavily biased toward a dominant modality, resulting in brittle cross-modal reasoning. We introduce C$^3$PO, a benchmark of 3,404 samples spanning video, audio, image, and text, evaluating two abilities: information composition (fusing dispersed evidence) and counterfactual conflict (resolving deliberate contradictions). C$^3$PO's paired IC/CC structure and four-tier design enable targeted diagnosis of when and why cross-modal reasoning fails. Built through a fully automatic pipeline using 25 logically grounded templates, C$^3$PO reveals that while humans achieve 88.64% accuracy, the best model (Gemini-3.1-Pro) reaches only 73.17%, with open-source models collapsing under conflict. Through attention probes, we find 86-95% of failures stem from modality dominance: models commit to one modality while ignoring contradictory evidence, concentrating 87-95% of attention on text. Mid-layer attention entropy predicts correctness-sustained exploration succeeds, premature collapse fails. The 56-point accuracy gap between equally complex templates reveals that performance depends on modalities' structural roles in conflict resolution, not combinations. These findings show multimodal perception does not guarantee robust reasoning; architectures must enable sustained cross-modal attention to avoid premature
Rakshanda Hassan Abhinandan, John Galeotti, Deva Ramanan +1cs.CV cs.AI cs.LG eess.IV
Where should the question go in a vision-language model (VLM) prompt: before the image or after it? Intuition says before: knowing what is asked should tell the model where to look. Yet across visual question answering benchmarks, question-first prompting consistently underperforms the image-first ordering recommended for frontier VLMs, a phenomenon we term the question-first paradox. We trace this paradox to a conflict between two stages of VLM computation. Logit-lens and attention probes show that question-first prompting steers perception, shifting image patch representations toward question-relevant concepts. But downstream, stranded behind hundreds of image tokens, the question is barely attended by the answer token, which instead commits to image-driven, often wrong answers. Causal attention knockout confirms that the answer reads the question only when it follows the image. This diagnosis yields a training-free fix: question echoing, restating the question on both sides of the image so one copy steers perception while the other is available at answer time. A similar division of labor appears in a fifty-year-old finding on human 'adjunct questions', where repeating a question before and after a passage improves comprehension. Echoing the image as well brings further gains by restoring the whole-image view otherwise lost by a causal decoder. The paradox holds across five open VLMs, costing up to 17.5 group-accuracy points. Echoed prompts recover most of the gap and, on NaturalBench and Winoground, surpass the best single-pass ordering by up to 19 group-accuracy points on Winoground, with no training, fine-tuning, or architecture change. The paradox reveals a tension between steering what a model sees and preserving access to what it was asked; echoing resolves this through prompt design. Project Page: https://rakshanda-cmu.github.io/ask-twice-look-twice/
Chain-of-Thought (CoT) prompting is widely used as a test-time scaling strategy for Vision-Language Models (VLMs), but it remains unclear what is extended when VLMs generate longer reasoning traces. We ask whether CoT requires continued access to image tokens, or whether it mainly operates over visual information already made available earlier in the forward pass. We introduce Visual Access Sweep, a causal intervention that masks attention from generated-token queries to image-token keys along layer depth and generation time, and define the Visual Access Boundary (VAB) as the minimal access region that preserves task accuracy. Across six model configurations from Qwen2.5-VL and InternVL3, both no-CoT direct answering and CoT prompting exhibit finite VABs. In Qwen2.5-VL-32B and InternVL3 at 14B and 38B scales, when CoT is evaluated against the no-CoT full-access target, its VAB layer differs from the no-CoT boundary by at most two layers, despite substantially longer generations. This suggests that CoT does not primarily improve performance by prolonging direct image-token access throughout the reasoning trace, but by extending language-side computation over image-derived hidden-state information. We further show that CoT gains are constrained by perceptual readout. CoT helps when the queried visual attribute can be reliably read out by the model, but not when that readout is unreliable. A symbolic-attribute oracle shows that CoT can improve counting once ground-truth attributes are supplied as text, while a single-object probe-vs-decode check shows that hard attributes can be linearly recoverable from hidden states yet difficult for the model itself to output. Together, these analyses place the bottleneck at readout rather than counting.
Multimodal large language models (MLLMs) generate responses autoregressively, integrating visual and linguistic information in an evolving context. Prior work on interpretability has focused on individual layers and circuits (where), leaving the token-level dynamics of multimodal computation during generation (when) underexplored. We address this gap and study attention shifts as per semantic role; tracking model attention to image, text, instruction, and previously generated tokens, One Token at a Time (OTaT). We introduce multimodal tasks that require explicit switching between visual and textual context within a single response. Across two mainstream model families and four open-weight MLLMs of varying sizes, we establish consistent patterns: attention to image peaks at tokens requiring image-derived information, instruction tokens are revisited during task transitions, and attention to previously generated tokens increases as the generation progresses. Causal attention blocking interventions validate the functional role of these trends. We profile model behavior under disrupted attention and observe responses falling back to language priors, or exhibiting cross-modal leakage, denial, or recovery. Finally, informed of the attention dynamics through our novel analysis, we propose a simple test-time intervention to boost attention to the relevant modality at the right time, significantly improving multimodal task performance.
Vision-language models (VLMs) often generate fluent but visually unsupported descriptions, especially by mentioning objects absent from the image. We propose QK Product Steering, a data-free, training-free, and zero-inference-cost weight edit for reducing object hallucination. The method directly edits the per-head query-key product, the operator that produces pre-softmax attention logits, by suppressing a small number of dominant singular modes in selected middle layers. The edited product is then mapped back to the query weights through a closed-form query-only update while keeping shared key weights fixed, making the edit compatible with grouped-query attention. We further decompose the QK product into symmetric and antisymmetric components to distinguish mutual content-similarity patterns from directional attention patterns. Across three GQA-based VLMs, QK Product Steering achieves an average relative CHAIR$_s$ reduction of $4.0\%$, while matched random-mode controls show negligible change. Interpretability ablations show that the hallucination signal is specific to dominant QK modes and is primarily localized to the symmetric mutual-attention channel. Overall, QK Product Steering offers a simple alternative to decoding-time mitigation, requiring no additional data, fine-tuning, or inference-time overhead while largely preserving general multimodal capability.
Large Audio Language Models (LALMs) achieve strong performance on a variety of audio understanding tasks but continue to struggle with temporal reasoning, a fundamental capability central to human auditory perception. Understanding the causes of these failures remains challenging as existing benchmarks report performance gaps without probing underlying mechanisms. To address this, we introduce a benchmark with 1,657 questions across three foundational tasks designed specifically for mechanistic analysis. Examining model outputs across varying input settings (behavioral analysis) reveals that models often under-utilize audio when textual cues are available. We also provide the first causal mechanistic analysis of temporal reasoning failures in LALMs. Comparing attention upweighting against scaling, we find that redistributing attention across audio tokens is more effective than increasing audio attention. Targeting task-relevant tokens yields further gains. These findings suggest that modality imbalance alone cannot explain failures. Attention scaling at bottleneck layers improves accuracy from 55.9% to 59.1% without fine-tuning, demonstrating a promising direction for future work.
Logan Mann, Yi Xia, Ajit Saravanan +6cs.CV cs.AI cs.CL cs.LG
Multimodal Foundation Models are increasingly used as reasoning agents, making reliability, knowing when a model may hallucinate, critical. A common intuition, which we call the Attention-Confidence Assumption, holds that reliability follows from "structural" visual perception: tight attention on relevant regions should signal a trustworthy answer, while scattered attention signals confusion. We challenge this through the VLM Reliability Probe (VRP), a systematic cross-family study of reliability signals in contemporary Vision-Language Models (VLMs). We introduce structural-attention metrics, cluster counts (C_k) and spatial entropy (H_s), to quantify the visual encoder's gaze, and track its evolution (Delta H_s) across layers. This reveals a "Symbolic Detachment": models often "Early Lock" visual features only to diffuse attention later, severing early perception from final generation. Contrary to the grounding hypothesis, we find a "Cluster Failure": spatial attention has near-zero correlation (R approx 0.001) with accuracy. Instead, reliability is a phenomenon of generation dynamics and internal-state distributions. Self-Consistency, the agreement rate across sampled reasoning paths, is the dominant predictor of truth (R = 0.429). Scaling causal interventions exposes a sharp architectural divergence: LLaVA locks its prediction in a fragile late-stage bottleneck, whereas PaliGemma and Qwen2-VL distribute reliability globally, staying resilient even when ~50% or more of their most predictive layer is destroyed. For current VLMs, reliability signals are detached from visual grounding maps and are best inferred from generation-time dynamics and hidden-state probes.
Visual Text Comprehension (VTC) renders text into images for a vision-language model (VLM) to read, sidestepping LLM context-window limits and powering applications from long-page OCR to multi-page memory QA. Yet existing VTC pipelines treat rendering and layout as a fixed, content-agnostic preprocessing step and offer little mechanistic understanding of how VLMs internally process visualized text. Through a focused empirical study on VTC QA tasks, we reveal that VLMs exhibit a localization-without-utilization regime: evidence-localizing attention emerges sharply in the middle-to-late layers and is largely decoupled from answer correctness, yet simply enlarging the localized spans on the rendered page recovers a large fraction of the failures. Building on these observations, we propose AGAR (Attention-Guided Adaptive Rendering), a training-free, model-agnostic method that leverages a VLM's own middle-to-late layer attention to identify the top-K important visual patches, maps them back to word spans, and re-renders the page with those spans enlarged before re-inferring the answer. Extensive experiments across nine VTC benchmarks (short-form, long-context, and multi-page memory QA) and four VLM backbones show that AGAR (i)consistently improves off-the-shelf VLMs as a plug-and-play enhancement, (ii)composes with VLM post-training to yield further gains, and (iii)remains robust under both visual- and text-side input degradation.
Multimodal LLMs are increasingly used to assist scientific peer review, where a core requirement is verifying whether claims in a paper are supported by its evidence. Prior work has shown that models perform substantially better at this task when the evidence is a table than when it is a chart of the same underlying data. This raises the question of whether models fail to extract information from charts, or do they extract it but fail to use it when forming their prediction? We study this question through layer-wise linear probing and attention analysis on three open-weight VLMs over table and chart evidence, representing the same underlying data. We find consistent evidence for the latter. Chart information is encoded in the models' intermediate representations but does not reach the prediction position, a gap that is absent for tables and holds across all conditions tested. Attention analysis further reveals that this disconnect takes two architecturally distinct forms across model families. These findings reframe the table-chart gap as a failure of how encoded visual information is routed at prediction time, rather than a failure of encoding itself.
Video temporal grounding (VTG), which localizes the start and end times of a queried event in an untrimmed video, is a key test of whether multimodal large language models (MLLMs) understand not only what happens but also when it happens. Although modern MLLMs describe video content fluently, their timestamp predictions remain unreliable, while existing remedies either require costly post-training on temporal annotations or rely on coarse training-free heuristics. In this work, we probe the cross-modal attention of MLLMs and uncover a perception-generation gap. Our key finding is that MLLMs often know the target interval during prefill, but lose this signal when generating the final answer. In the prefill stage, a sparse set of attention heads, which we call \emph{Temporal Grounding Heads} (TG-Heads), concentrates query-to-video attention on the ground-truth interval. During autoregressive decoding, however, the answer tokens shift attention away from this interval toward visually salient but query-irrelevant segments. This observation motivates an inference-time read-then-regenerate framework. We first convert TG-Head prefill attention into a debiased frame-level relevance signal and extract the high-attention interval it highlights. We then re-invoke the MLLM with visual context restricted to this interval, using video cropping or attention masking to suppress distractors. Without parameter updates and architectural changes, our framework consistently improves MiMo-VL-7B, Qwen3-VL-8B, and TimeLens-8B on three VTG benchmarks, with gains of up to +3.5 mIoU. The project website can be found at https://ddz16.github.io/mllmsknowwhen.github.io/.
Vision-Language Models (VLMs) have recently demonstrated remarkable capabilities in visual understanding and reasoning, but they also impose significant computational burdens due to long visual sequence inputs. Recent works address this issue by pruning unimportant visual tokens, achieving substantial computational reduction while maintaining model performance. The core of token pruning lies in determining token importance, with current approaches primarily relying on attention scores from vision encoders or Large Language Models (LLMs). In this paper, we analyze the effectiveness of attention mechanisms in both vision encoders and LLMs. We find that vision encoders suffer from attention sink, leading to poor focus on informative foreground regions, while in LLMs, although prior studies have identified attention bias toward token positions, text-to-vision attention demonstrates resistance to this bias and enables effective pruning guidance in middle layers. Based on these observations, we propose LearnPruner, a two-stage token pruning framework that first removes redundant vision tokens via a learnable pruning module after the vision encoder, then retains only task-relevant tokens in the LLM's middle layer. Experimental results show that our LearnPruner can preserve approximately 95% of the original performance while using only 5.5% of vision tokens, and achieve 3.2$\times$ inference acceleration, demonstrating a superior accuracy-efficiency trade-off.