A study led by Dr. Kevin Jablonka, associate member of HIPOLE Jena and junior research group leader at Friedrich Schiller University Jena, reveals a fundamental limitation of AI models in material science: While language models can identify which atoms a material consists of, they struggle to capture how those atoms are arranged in space. Since properties such as hardness and electrical conductivity often depend on atomic geometry (e.g., diamond vs. graphite), predictions made by these models can be unreliable for many applications. Therefore, specialized AI models designed to process spatial relationships are the better choice in this case.
The study was published in Nature Computational Science and was conducted in collaboration between Friedrich Schiller University Jena, HIPOLE Jena, and Intel Labs in the United States.
The full article can be found here: https://www.nature.com/articles/s43588-026-01045-2
Furthermore, you can read about the findings and their implications in the HZB article: Blind spot for the arrangement of atoms – Helmholtz-Zentrum Berlin (HZB)