Armstrong L1, Johnson T
1Waipapa Taumata Rau | University of Auckland, Auckland, New Zealand
Biography:
Toby Johnson works as an eResearch Engagement Specialist at the Centre for eResearch, Waipapa Taumata Rau | University of Auckland. He received his PhD in Psychology from the University of Exeter where he investigated computational models of learning. At the Centre for eResearch he provides digital research and AI skills training and assists researchers in accessing compute resources, including the Secure Research Environments.
Abstract:
Large language models (LLMs) have been rapidly adopted by researchers, often through accessible, no-code chatbot interfaces. However, the greater research benefits lie in using LLMs programmatically: a documented, re-runnable pipeline allows researchers to process data at scale and produce reproducible results. Ad hoc chatbot interactions simply cannot offer this level of rigor. Our goal was to take researchers beyond the chatbot and teach them to apply LLMs programmatically in their own workflows.
At the Centre for eResearch, Waipapa Taumata Rau | University of Auckland, our training initially focused entirely on the programmatic use of LLMs. However, we found this steep learning curve was often a barrier for researchers. This forced us to reconsider how to balance chatbot utility against programmatic power, a shift that came with real trade-offs in scale and reproducibility. How we navigated this tension, and the hybrid approach we ultimately landed on, is the heart of this talk. We argue that while chatbots serve as a valuable on-ramp, programmatic execution remains the destination for robust research.
Beyond this technical balance, we reflect on identifying community needs, refining materials through researcher feedback, and linking training to institutional platforms. Attendees will leave with practical, transferable guidance for designing AI training at their own institutions, and a framework for meeting researchers where they are.
