Haniyeh Ehsani Oskouie

MLLMs generate critiques that are more verbose and less selective than human reviewers, raising questions about the validity of traditional evaluation metrics.

Triangular consistency can universally enhance optical flow learning, yielding consistent performance gains across diverse training settings without additional annotations.

LLMs can't reliably predict scientific experiment outcomes, and more worryingly, they have no idea when they're wrong, unlike human experts whose accuracy skyrockets when they feel confident.