---
license: mit
task_categories:
- image-text-to-text
---

Paper: Hidden in Plain Sight: Probing Implicit Reasoning in Multimodal Language Models

Paper: https://arxiv.org/abs/2506.00258

Website: https://jackie-2000.github.io/iReason.github.io/

GitHub: https://github.com/eric-ai-lab/iReason

iReason: Designed to probe MLLMs’ implicit reasoning by evaluating their ability to detect subtle flaws in seemingly valid instructions. Covers 643 real-world scenarios spanning four failure types—object absence, ambiguity, contradiction, and infeasibility—challenging models to surface issues that are “hidden in plain sight” without explicit cues.

For more detailed instructions on inference, please refer to our GitHub repo.