Probing the limitations of multimodal language models for chemistry and materials research