A user uploads a PDF of a paper to a large model, which scores the paper and provides corresponding reasonable reasons. After the user adds perturbations to the illustrations in the paper and uploads it again to the same large model, the model gives the paper a higher score. See more