Senses Wide Shut: A Representation-Action Gap in Omnimodal LLMs
Trung Nguyen Quang, Yiming Gao, Fanyi Pu, Kaichen Zhang, Shuo Sun, Ziwei Liu
Why It Matters
What makes this one worth your time
Understanding the limitations of omnimodal models in grounding can lead to improvements in their design and application, particularly in real-world scenarios where accurate perception is crucial.
This research highlights a critical gap in omnimodal LLMs' ability to reconcile sensory input with textual claims.
Summary
The paper introduces a benchmark for evaluating omnimodal large language models' ability to detect contradictions between textual claims and their sensory inputs, revealing a significant gap in their grounding capabilities.
Key contributions
- Development of the IMAVB benchmark for testing conflict detection in omnimodal LLMs.
- Identification of the Representation-Action Gap in model outputs.
- Demonstration of the effectiveness of PGLA in enhancing rejection behavior.
Notable insights
- The study identifies a modality-asymmetric performance in grounding, with audio grounding being less effective than vision.
- The introduction of a probe-guided logit adjustment (PGLA) offers a novel approach to improving model rejection behavior.
Possible limitations
- Not stated in the abstract.
Abstract
arXiv:2605.13737v1 Announce Type: new Abstract: When an omnimodal large language model accepts a question whose textual premise contradicts what it actually sees or hears, does the failure lie in perception or in action? Recent omnimodal models are positioned as perception-grounded agents that jointly process video, audio, and text, yet a basic form of grounding remains untested: catching a textual claim that conflicts with the model's own sensory input. We introduce IMAVB, a curated 500-clip benchmark of long-form movies with a 2x2 design crossing target modality (vision, audio) and premise condition (standard, misleading), which lets us measure conflict detection separately from ordinary multimodal comprehension. Across eight open-source omnimodal LLMs and Gemini 3.1 Pro, we document a Representation-Action Gap: hidden states reliably encode premise-perception mismatches even when the same models almost never reject the false claim in their outputs. Behaviorally, models fall into two failure modes: under-rejection, in which they answer misleading questions as if the false premise were true; and over-rejection, in which they reject more often but also reject standard questions, sacrificing ordinary comprehension accuracy. The gap is modality-asymmetric (audio grounding underperforms vision) and prompt-resistant across seven variants. As an initial diagnostic intervention, a probe-guided logit adjustment (PGLA) re-injects the encoded mismatch signal into decoding and consistently improves rejection behavior. Together, these results suggest the bottleneck for omnimodal grounding lies in translation, not perception.