En este curso me he lanzado a crear un Gem y un NotebookLM para la asignatura Topología I del Grado en Matemáticas, usando la cuenta go-ugr-es. En esta charla quiero contar cómo los estoy desarrollando, aunque quizá lo más importante sea la lista de errores cometidos. Haremos un Gem y un NotebookLM muy básicos in situ.

Can large language models improve feedback on undergraduate proof writing without lowering standards? I present a practical way to guide an LLM so its comments are focused, consistent, and aligned with course aims—while academic judgement stays with the instructor. In a pilot with first-year work, this approach surfaced common misconceptions early and shortened feedback cycles. I’ll outline the idea, note what helped and what failed, and discuss where AI can responsibly sit alongside human marking.