Practical course about LLMs for advanced BA students
Theory & practical sessions with high involvement
6 credits, pass/fail
Contents¶
Summary¶
When they were first introduced, LLMs were mainly seen as seq2seq models exclusively for text generation: input text in — output text out. However, as LLMs evolved, their application domain quickly expanded, and they are now used for purposes few could have imagined: from powering websites and automating business processes to drug discovery. The central goals of this course are:
to provide a comprehensive overview of how the LLMs are used in the real world;
to teach you the basics of implementing LLM-based pipelines for various purposes.
While the course will mostly focus on the practical uses of LLMs, it will also briefly touch on ethical, societal, and broader considerations.
The course contents, slides, tasks, reading, etc. will be hosted in this webbook. Moodle will be used for administrative stuff.
Contents Overview¶
The course will consist of the intro (Weeks 1-2), the core (Weeks 3-9), and the project work (Weeks 10-14). You can see the schedule in Topics Overview.
The into sessions will introduce the course and provide a recap on LLMs --- both ontological and technical aspects.
The core sessions will mostly consist of practical sessions, where you will be mastering building LLM-based single- and multi-agent systems. By the end of the core block, you will have created two toy applications as a preparation for the final project. Additionally, a lecture about the role of AI in recent years will be offered, followed by debates.
Finally, you will build and present an advanced multi-agent system as your final project.
Formats¶
The course will feature both theoretical and practical sessions.
Theory:
Lectures. At the lectures, we will mostly discuss applications of LLMs in different fields (on the examples of existing solutions), how the respective solutions are built and how they function, what their limitations are etc. You are highly encouraged to engage in discussions and arguments, constructively criticize the solutions and suggest your improvements and edits.
Practice:
Labs. In labs, we will practice technical skills by implementing toy LLM-based solutions.
Debates. The debates will be conducted on material of the sessions dedicated to philosophical / ethical etc. considerations (weeks 1 and 13). In the debates, a certain motion (position) will be debated over by two teams: one will support it (the affirmative) and one will oppose it (the negative). See Debates for more details.
Project Presentations. This is a usual presentation that will cover the implementation of your final project. The presentation will have to focus on justification of the technical solutions that will be used in the application.
Apart from the content-heavy sessions, there will be a few administrative meeting in which we will discuss questions, issues, next steps etc.
Target Audience¶
The course is intended for advanced BA students of computational linguistics, who have interest in practical applications of LLMs. MA students cannot get credits this course. Practical experience with programming in Python and applying LLMs are a prerequisite (more formally, you should have completed ISCL-BA06, ISCL-BA07, and ISCL-BA08 successfully). Knowledge in LangChain and LangGraph is encouraged but not necessary.
Formalia¶
This is a pass/fail course for 6 credits (there is no 9 credits option). The course is counted towards ISCL-BA-11 (Spezialisierungsmodul).
Each student will have to participate in one debate (any side) and complete and present the final project to pass the course. There will be no exam, and no in-person presence is required.