In this graduate-level course, we will learn how large language models are trained, evaluated, and adapted from a human-centered perspective. We will begin with an introduction to the transformer architecture and autoregressive language modeling. The first few weeks of the course will focus on pre-training, with an emphasis on training data selection, model memorization, mid-training. Next we will focus on post-training techniques and applications. Finally we will explore topics on Generative AI and Fair Use, AI detection, Impacts of AI on mental health and wellbeing and AI Agents. This course will consist of lectures (10% grade for attendance), readings and discussions (10% grade on class participation), student presentations on selected papers (20% of the grade), along with a in-class midterm exam (15% of the grade) and a final project (45% of the grade).
Prerequisites: Knowledge of LLMs. Have taken NLP or AI.
Logistics:
- Timing: Tuesday/Thursday 12:30-1:50 PM
- Location: NCS 120
- Device Policy: Use of Laptop during class is not allowed. Please engage with lectures. Using ipad for taking notes is fine. If you must need laptop or special accomodation please email the instructor.
Tuhin Chakrabarty
