{"version":"1","article":{"id":"llms-coordinate-data-material-prediction-limitations-2026","slug":"llms-coordinate-data-material-prediction-limitations-2026","language":"en","url":"https://haberpaper.com/en/news/llms-coordinate-data-material-prediction-limitations-2026","api_url":"https://haberpaper.com/api/v1/articles/llms-coordinate-data-material-prediction-limitations-2026?lang=en","title":"Language models can miss atomic geometry in materials property prediction","summary":"Research co-authored by Intel Labs reports that language models learn composition labels better than spatial coordinates. Geometry-focused graph networks performed better on these tasks.","category":"business","region":"world","topics":["ai"],"published_at":"2026-10-13T20:11:52+03:00","modified_at":"2026-10-13T20:11:52+03:00","publication_basis":"publication-record","event_period":"2026-10-12","event_period_basis":"event-or-reference-period","sources":[{"name":"Nature Computational Science, Less can be more for predicting properties with large language models","url":"https://www.nature.com/articles/s43588-026-01045-2","checkedAt":"2026-10-13","publishedAt":"2026-10-12"}],"translations":[{"language":"tr","url":"https://haberpaper.com/tr/news/llms-coordinate-data-material-prediction-limitations-2026","api_url":"https://haberpaper.com/api/v1/articles/llms-coordinate-data-material-prediction-limitations-2026?lang=tr"},{"language":"en","url":"https://haberpaper.com/en/news/llms-coordinate-data-material-prediction-limitations-2026","api_url":"https://haberpaper.com/api/v1/articles/llms-coordinate-data-material-prediction-limitations-2026?lang=en"},{"language":"ar","url":"https://haberpaper.com/ar/news/llms-coordinate-data-material-prediction-limitations-2026","api_url":"https://haberpaper.com/api/v1/articles/llms-coordinate-data-material-prediction-limitations-2026?lang=ar"},{"language":"fr","url":"https://haberpaper.com/fr/news/llms-coordinate-data-material-prediction-limitations-2026","api_url":"https://haberpaper.com/api/v1/articles/llms-coordinate-data-material-prediction-limitations-2026?lang=fr"},{"language":"es","url":"https://haberpaper.com/es/news/llms-coordinate-data-material-prediction-limitations-2026","api_url":"https://haberpaper.com/api/v1/articles/llms-coordinate-data-material-prediction-limitations-2026?lang=es"}],"data_types":[],"data":null,"data_compilations":[],"data_downloads":[],"content":{"paragraphs":["A study published on 12 October in Nature Computational Science examined how large language models use positional and categorical information in scientific data. Co-authored by researchers at Friedrich Schiller University Jena and Santiago Miret of Intel Labs, it combines controlled synthetic-data experiments with materials-property benchmarks. The authors report that language models can learn composition labels such as atom types but systematically struggle with coordinates describing how atoms are arranged in space.","The gap persisted in the tested models up to 70 billion parameters and datasets of up to two million structures. Graph neural networks that explicitly model geometry outperformed language-model approaches on structure-dependent material properties. The comparison suggests language models are not suitable for every scientific prediction task; it does not report a newly produced material or laboratory validation."],"key_points":[],"sections":[]},"text":"Research co-authored by Intel Labs reports that language models learn composition labels better than spatial coordinates. Geometry-focused graph networks performed better on these tasks.\n\nA study published on 12 October in Nature Computational Science examined how large language models use positional and categorical information in scientific data. Co-authored by researchers at Friedrich Schiller University Jena and Santiago Miret of Intel Labs, it combines controlled synthetic-data experiments with materials-property benchmarks. The authors report that language models can learn composition labels such as atom types but systematically struggle with coordinates describing how atoms are arranged in space.\n\nThe gap persisted in the tested models up to 70 billion parameters and datasets of up to two million structures. Graph neural networks that explicitly model geometry outperformed language-model approaches on structure-dependent material properties. The comparison suggests language models are not suitable for every scientific prediction task; it does not report a newly produced material or laboratory validation.","corrections":[],"editorial_illustration":{"kind":"symbolic-editorial-illustration","documentary":false,"scope":"section","asset_id":"business","url":"https://haberpaper.com/assets/illustrations/business-1536.webp","width":1536,"height":1024,"alt":"Symbolic industrial workshop with machinery, a warehouse and a crane.","language":"en"},"section_illustration":{"kind":"symbolic-editorial-illustration","documentary":false,"scope":"section","asset_id":"business","url":"https://haberpaper.com/assets/illustrations/business-1536.webp","width":1536,"height":1024,"alt":"Symbolic industrial workshop with machinery, a warehouse and a crane.","language":"en"}}}