Course overview: LLMs, agents & RL#
This course taught as a block seminar over the course of one week at Uni Tübingen will look at different approaches and aspects of agent modeling, i.e., building agents that complete various tasks in an environment in a goal-directed way. The course will cover approaches from classical cognitive science (e.g., discuss cognitive architectures like SOAR), RL (e.g., agents playing Atari-style games), as well as recent Large Language Model-based (LLM) agents (e.g., ToolFormer, or Generative Agents). The course will provide a high-level recap of LLMs, but familiarity with NLP and LLMs is recommended for this course. Intermediate programming skills are strongly encouraged.
Intended audience#
The course is intended for advanced bachelor students and master students interested in interdisciplinary topics in computational cognitive science and AI. The course is intended to be a mix of lectures, paper discussions and hands-on tasks, as provided through this webbook.
Schedule#
The overview of overall topics by day is below.
Introduction: what are agents?
Cognitive architectures
RL
LLM agents
Further materials#
Materials for this course are inspired by the following courses:
Outputs from a certain language model inspired some aspects of the materials, too.
For more in-depth recap or introduction of LLMs, please refer to, e.g., this webbook I co-authored and the resources listed therein.