COSC 89.35 Human-Centered Approaches to Large Language Models: Design, Methods, and Evaluation

The goal of this course is to prepare students to critically assess and design LLMs for real-world applications by combining technical and human-centered perspectives. This course explores the foundations, design, and evaluation of large language models (LLMs) through a human-centered lens. The course covers core concepts such as pretraining, fine-tuning, and instruction tuning, as well as the role and implications of data sources. Additionally, the course examines inference methods, including in-context learning, retrieval-augmented generation (RAG), and using LLMs as agents. A key focus is on evaluating LLMs for alignment, hallucination, fairness, bias, toxicity, reliability, stability, safety, and robustness. Students will analyze how considering human-centered approaches in the design, development and evaluation of LLM-based solutions are essential for responsible and ethical AI. Course activities will include reviewing research articles, student-led presentations, in-class and canvas discussions, writing short reaction papers, and a course project.

Degree Requirement Attributes

Dist:TAS

The Timetable of Class Meetings contains the most up-to-date information about a course. It includes not only the meeting time and instructor, but also its official distributive and/or world culture designation. This information supersedes any information you may see elsewhere, to include what may appear in this ORC/Catalog or on a department/program website. Note that course attributes may change term to term therefore those in effect are those (only) during the term in which you enroll in the course.

Department-Specific Course Categories

Computer Science