The paper shows that controlled visual in‑context learning can help vision‑language models (VLMs) overcome errors caused by training priors. By providing demonstrations, the models focus more on visual evidence, yielding significant accuracy gains on counterfactual examples while leaving prior‑aligned performance largely unchanged. This effect is observed across five VLM families and five vision‑centric tasks, indicating that many VLMs possess latent visual grounding abilities that standard tests may miss.
By Kun Wang, Xindi Wu, Sanghyuk Chun, Olga Russakovsky, Esin Tureci
The paper investigates why multimodal large language models (MLLMs) struggle with vision‑centric tasks when visual evidence conflicts with pretrained language knowledge. Using image reconstruction and a new WhatIfVis benchmark, the authors show that MLLMs preserve coarse‑grained visual attributes but fail to consistently use them, and that supervised fine‑tuning and activation patching can improve controllability of visual context sensitivity. The study demonstrates that the main bottleneck lies in the models’ inability to reliably regulate their reliance on visual evidence rather than in visual perception itself.
By Jiaang Li, Chengzu Li, Zhaochong An, Yifei Yuan, Xi Liu, Serge Belongie, V\'esteinn Sn{\ae}bjarnarson
arXiv:2605.05593v2 Announce Type: replace
Abstract: Despite the remarkable success of Multimodal Large Language Models (MLLMs) across diverse tasks, the internal mechanisms governing how they encode...
By Zehao Deng, Tianjie Ju, Zheng Wu, Liangbo He, Jun Lan, Huijia Zhu, Weiqiang Wang, Zhuosheng Zhang
CounterCount is a diagnostic dataset that pairs factual and counterfactual images to test whether Vision‑Language Models (VLMs) rely on visual evidence or on world knowledge when counting attributes. Humans achieve near‑perfect accuracy on both types of images, whereas VLMs drop from 91.6% to 49.5% on counterfactuals, often defaulting to canonical counts. The study shows that VLMs exhibit a System‑1‑like bias, underutilize visual evidence, and that targeted input manipulations can mitigate this bias.
By Reem Alzahrani, Bushra Bin Hemid, Hassan Alshanqiti, Zaid Alyafeai, Paul Gavrikov, Bernard Ghanem, Abdelrahman Eldesokey
arXiv:2606. 19297v1 Announce Type: new Abstract: Embodied Vision-Language-Action (VLA) models are typically obtained by fine-tuning powerful pretrained VLMs on robotics data, yet it is unclear how much commonsense and factual knowledge they retain after adaptation.
By Nikita Kachaev, Andrey Moskalenko, Matvey Skripkin, Nikita Kurlaev, Daria Pugacheva, Albina Burlova, Mikhail Kolosov, Denis Shepelev, Andrey Kuznetsov, Elena Tutubalina, Aleksandr I. Panov, Alexey K. Kovalev, Vlad Shakhuro
arXiv:2604. 14888v3 Announce Type: replace-cross Abstract: Recent advances in vision language models (VLMs) offer reasoning capabilities, yet how these unfold and integrate visual and textual information remains unclear.
By Danae S\'anchez Villegas, Samuel Lewis-Lim, Nikolaos Aletras, Desmond Elliott
The study investigates whether attention weights in Vision‑Language Models (VLMs) accurately reflect model reasoning for visual inputs. Using causal perturbation analysis, it identifies three distinct processing modes—Faithful‑Sufficient, Faithful‑Distributed, and Non‑Focal—indicating heterogeneous visual attention faithfulness. The research also shows that human‑annotated ground‑truth regions align with model attention in only about 60% of cases, highlighting a systematic divergence between model visual reliance and human intuition across VQA, document, and chart tasks.
By Xurui Song, Weishi Wang, Zhongqi Yue, Kuluhan Binici, Tao Bai, Hongxin Shao, Daniel Dahlmeier, Jun Luo
VKnowU is a benchmark that tests multimodal large language models (MLLMs) on their grasp of visual knowledge—intuitive, human-like understanding of physical and social principles in videos. The benchmark contains 1,680 questions across 1,249 videos, covering eight core types of visual knowledge, and shows that current state‑of‑the‑art MLLMs still lag behind human performance, especially on world‑centric tasks. To address this gap, the authors release VKnowQA and VideoKnow+, a baseline model that incorporates visual knowledge via a See‑Think‑Answer framework and reinforcement learning, improving performance on VKnowU and related datasets.
By Tianxiang Jiang, Sheng Xia, Yicheng Xu, Linquan Wu, Xiangyu Zeng, Limin Wang, Yu Qiao, Yi Wang
arXiv:2606. 06890v1 Announce Type: cross Abstract: Vision-Language Models (VLMs) frequently rely on language priors, producing confident answers that are weakly grounded in visual evidence.
By Runyu Zhou, Qi Zhang, Qixun Wang, Yisen Wang
Embodied Vision-Language-Action (VLA) models are typically obtained by fine-tuning powerful pretrained VLMs on robotics data, yet it is unclear how much commonsense and factual knowledge they retain after adaptation. Failures on knowledge-sensitive tasks are ambiguous, conflating missing knowledge with poor generalization of low-level control.
arXiv:2606. 19965v1 Announce Type: cross Abstract: Multimodal large language models (MLLMs) are increasingly expected to act on visual information, yet the same scene may require different actions under different task contexts.
By Yihao Wang, Zijian He, Jie Ren, Keze Wang
arXiv:2606. 17389v1 Announce Type: cross Abstract: Multimodal Foundation Models are increasingly used as reasoning agents, making reliability, knowing when a model may hallucinate, critical.
By Logan Mann, Yi Xia, Ajit Saravanan, Ishan Dave, Saadullah Ismail, Shikhar Shiromani, Emily Huang, Ruizhe Li, Kevin Zhu