Encoded Not Routet_news

Multimodal LLMs are much better at verifying claims from scientific tables than equivalent charts. Our new preprint investigates why🕵️‍♂️: do models fail to extract chart information, or fail to use it effectively once extracted? Check it out!