About This Course

Agentic Large Language Models is a graduate course offered by the Leiden Institute of Advanced Computer Science (LIACS) at Leiden University, part of the Master’s programs Computer Science (AI track) and Creative Intelligence & Technology.

Course Description

Agentic LLMs are LLMs that (1) reason, (2) act, and (3) interact. After an introduction to LLM architecture and training procedures, the course is organized around these three categories:

  • Reasoning, reflection, and retrieval — improving the decision making of a single LLM.
  • Action models, robots, and tools — turning LLMs into useful assistants that act in the world.
  • Multi-agent systems — collaborative task solving and simulated interaction to study emergent social behavior.

Throughout the course we discuss applications of agentic LLMs in domains such as medical diagnosis, logistics, and financial market analysis. We also look at how self-reflective agents that play roles and interact with one another can augment scientific research itself, and how agentic LLMs offer a way around LLMs running out of fresh training data: inference-time interaction can generate new training signal, so that models can keep learning without ever-larger datasets. Agentic LLM assistants that act in the real world also carry real risk, and we discuss both the risks and the societal benefits throughout the course.

Course Structure

  • Lectures: Weekly sessions covering the theory and case studies from the course textbook, Agentic Large Language Models (Van Duijn, Van der Meer, Plaat & Van Stein).
  • Case studies / assignments: Four hands-on programming case studies, each spanning about three weeks — see the Schedule for exact dates.

See the Syllabus for prerequisites, learning outcomes, grading, and the full breakdown of parts, chapters, and assignments.

Teaching Team

The course is taught by its four textbook authors, all researchers at LIACS, Leiden University:

Max van Duijn — Assistant professor at LIACS and PI of the Social Intelligence Modelling (SIM) lab. His research combines cognitive science, linguistics, and AI to study social intelligence, in particular empathy, perspective-taking, and Theory of Mind. He co-founded the Creative Intelligence Lab and chairs The Young Academy (De Jonge Akademie).

Michiel van der Meer — Postdoctoral researcher at LIACS working on AI, natural language processing, and hybrid intelligence. His research develops technology for human-AI collaboration, addressing subjectivity, uncertainty, and diversity of perspectives in language processing.

Aske Plaat — Professor of Artificial Intelligence at Leiden University. He heads the Reinforcement Learning Group and the Sustainable Energy Learning Lab, and is the author of two textbooks on Deep Reinforcement Learning.

Niki van Stein — Associate professor at LIACS, leading the Explainable AI (XAI) group within the Natural Computing cluster. Her research focuses on integrating explainability into AI models for predictive maintenance, engineering design, and automated algorithm design.

Contact

For questions about the course, contact the lecturers at a.plaat@liacs.leidenuniv.nl, or visit the LIACS website.