Epoch’s AI use in research explorer tracks large language model (LLM) use acknowledgments in research papers across physics, statistics, and mathematics preprints on arXiv and biology preprints on bioRxiv. We also track acknowledgments of specialized AI tool use in biology.

We search the full text of every preprint posted on arXiv from January 2023 to the present and bioRxiv from January 2020 to the present for AI and LLM keywords. We then use GPT-5.6 Sol to read the passages containing the keywords and the paper’s acknowledgment and declaration sections, and decide whether the authors acknowledge using an LLM. We use the OpenAI Decisions API to categorize the LLM use acknowledgments into five use cases: research, code and computation, writing, literature review, and purpose not stated.

This explorer uses preprints as a sample of research papers. In math, over 95% of research papers published in top journals are first posted to arXiv. In other fields, coverage varies. We estimate that about 80% of physics papers, 70% of statistics papers, and 30% of biology papers published in top journals first appeared as preprints on arXiv or bioRxiv. Preprints may not be a representative sample of overall papers in a field: for example, we found that about 36% of published computational biology papers were first posted as preprints, whereas only about 25% of published ecology papers were first posted as preprints.

Our methodology is adapted from the methodology used in our AI use in math research explorer.