Working responsibly with LLMs

Slides from Part I of a workshop on the practical use of large language models (LLMs) and agentic AI systems in scientific work at the Helmholtz Centre for Environmental Research (UFZ) for our research unit Chemicals in the Environment - CITE. This workshop part, “Introduction to LLMs in Research,” introduces non-computer-science researchers, research assistants, and managers to the capabilities and limitations of contemporary LLMs (e.g., GPT-5.1, Gemini) and their use in everyday research and administrative workflows. The focus is on realistic expectations, responsible use, and concrete practices that help participants work more efficiently while maintaining scientific integrity. The slides cover: A very brief, intuitive introduction to LLMs and agentic AI (goal–workflow perspective) Realistic vs. unrealistic expectations for today’s models in scientific contexts Core prompting patterns (role–task–context–constraints) and iterative refinement How to include additional context via text, documents, web links, and images Basic guidance on responsible use: verification of outputs, data protection considerations, transparency about AI assistance These materials are intended as a reusable teaching resource for institutions that wish to train researchers and support staff in the effective and responsible use of LLM-based tools for writing, communication, and light analytical tasks. Note: This upload only contains the slides for Part I (LLM basics, expectations, prompting, responsible use). Slides and demos related to institution-internal systems presented in later parts of the workshop are not included in this Zenodo record.