Search papers, labs, and topics across Lattice.
This study investigates the limitations of Multimodal Large Language Models (MLLMs) in vision-centric tasks, particularly when visual information conflicts with pretrained knowledge. By employing two diagnostic paradigms鈥攊mage reconstruction and multimodal context sensitivity measurement鈥攖he authors reveal that while MLLMs preserve coarse-grained visual evidence, they struggle with post-perceptual utilization, leading to inconsistent reliance on visual context. The introduction of the WhatIfVis benchmark and subsequent findings highlight that supervised fine-tuning can enhance controllability over visual context sensitivity, suggesting that the core issue lies not in visual encoding but in the models' ability to manage their reliance on visual inputs.
MLLMs can encode visual evidence but often fail to control their reliance on it, revealing a critical bottleneck in multimodal reasoning.
Multimodal Large Language Models (MLLMs) achieve strong performance by integrating visual inputs with the rich priors of pretrained language models. However, they often fail on vision-centric tasks, especially when visual evidence conflicts with pretrained knowledge. We explore these failures separately using two diagnostic paradigms: (1) probing whether visual information is available, via image reconstruction, and (2) measuring multimodal context sensitivity, the extent to which the model follows visual context versus the language prior. To support the second, we introduce the WhatIfVis, a benchmark spanning five coarse-grained dimensions (spatial-temporal, color, count, size, and weight) whose questions admit answers from either the image or the prior. Our analysis yields three findings: (i) Coarse-grained visual evidence is preserved, as these attributes can be reconstructed from the final-layer image tokens of frozen MLLMs. Failures on questions about these attributes therefore point to post-perceptual utilization, rather than to degraded visual encoding during perception. (ii) Even when explicitly instructed to use or ignore visual evidence, vanilla models (without supervised fine-tuning on the WhatIfVis) show unstable visual context sensitivity. Supervised fine-tuning (SFT) improves this controllability and generalizes across domains, and activation patching further localizes the vision-versus-prior trade-off at architecture-specific depths across all six models. (iii) The vision-versus-prior trade-off is controllable along a learned vector. Applying this steering vector, even without any intent instruction, improves controllability over the vanilla model. Together, these results relocate the bottleneck, indicating that for the coarse attributes we study, MLLMs encode the visual evidence but cannot reliably control their reliance on it.
ETH