aiqbench analysis
This arXiv paper proposes a diagnostic protocol that splits a vision-language model's errors into seeing, thinking, and using perception, without needing access to model weights or answer labels for the use measure. Its main finding is that models often answer from common priors rather than the image, and that asking the model to localize objects first can fix this when its boxes are accurate. For researchers and developers, the practical value is a way to direct test-time compute to the stage that is actually failing, though the results rest on two model families and six datasets and the code has not yet been released.
See, Think, Use: Diagnosing Perception and Reasoning Failures in VLMs
arXiv:2611.01467v1 Announce Type: cross Abstract: We present See, Think, Use, a protocol that traces each answer of a vision-language model to perception, reasoning or the use of perception. The diagnosis shows a developer which stage to improve and directs test-time compute to the stage that fails. Benchmarks score the final answer, and an answer that rests on the image earns the same credit as a guess or a memorized prior. The protocol asks the model two extra questions per item and needs no access to its weights. A localization probe asks for a box around every instance and measures seeing, and a text-replacement probe answers from a description and measures thinking. Use is the rate at which the answer follows the model's own perception across counterfactual image pairs. The rate needs no answer labels, because the question and its prior are the same on both images. On two model families and six datasets the protocol shows that the knob that helps follows the task family. Answers follow the model's perception except on images that contradict a common prior, and asking the model to localize first repairs this failure when its boxes are right. Allocating compute by the measured stages comes close to the best uniform policy at lower cost and beats a random allocation of the same compute wherever a knob raises the short stage. We will release the code, the prompts and the per-question outputs.
Abstract:We present See, Think, Use, a protocol that traces each answer of a vision-language model to perception, reasoning or the use of perception. The diagnosis shows a developer which stage to improve and directs test-time compute to the stage that fails. Benchmarks score the final answer, and an answer that rests on the image earns the same credit as a guess or a memorized prior. The protocol asks the model two extra questions per item and needs no access to its weights. A localization probe asks for a box around every instance and measures seeing, and a text-replacement probe answers from a description and measures thinking. Use is the rate at which the answer follows the model's own perception across counterfactual image pairs. The rate needs no answer labels, because the question and its prior are the same on both images. On two model families and six datasets the protocol shows that the knob that helps follows the task family. Answers follow the model's perception except on images that contradict a common prior, and asking the model to localize first repairs this failure when its boxes are right. Allocating compute by the measured stages comes close to the best uniform policy at lower cost and beats a random allocation of the same compute wherever a knob raises the short stage. We will release the code, the prompts and the per-question outputs.
Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
arXiv:2611.01467 [cs.CV]
(or arXiv:2611.01467v1 [cs.CV] for this version)
https://doi.org/10.48550/arXiv.2611.01467 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 335 words
Related coverage
Diagnosing and Mitigating Perception-Decision Misalignment in Omni-LLMs via Modality Subspace Activation
arXiv · Aug 19, 2026
ActiveArena: Benchmarking and Understanding Active Perception in Robotic Manipulation
arXiv · Sep 22, 2026
Reflect After Failure: Brain-Inspired Self-Evolving Agents for Embodied Spatial Intelligence
arXiv · Oct 15, 2026
Where MLLMs Fail and Why: Causal Task Decomposition for Capability Failure Diagnosis
arXiv · Oct 1, 2026