ART415/543
Monday 15:15 - 18:00
CFA 136
Reg# ART543 (#23052) / ART415 (#22205)
Top: John Fishetti in collaboration with GPT-2, 2022.
Bottom: Ruolin Chen, Xiaoxu Dong, Saiteja Gangavaram, Skrikar Gopaladinne,
Hemanth Gorla, Divya Kandukuri, Kathryn Korenblyum, Sean Mansfeld and Kelsey Rupe in collaboration with GPT-3.5 and DALL-E: Neural Recipe Box, 2023.
Overview
This advanced undergraduate and graduate course introduces students to the art and science of probing the behavior of Generative Artificial Intelligence (GenAI) systems through critical thinking and AI literacy.
Artificial Intelligence—particularly its generative branch GenAI—has become a site of cultural and technical transformation, reshaping how knowledge is produced and interpreted across disciplines. Unlike traditional software with deterministic outputs, GenAI systems can produce varied and context-dependent responses to the same input, making their study both technically and critically complex.
The course begins with a survey of early generative paradigms such as genetic algorithms before turning to the foundational mechanisms of modern Large Language Models. Core topics include transformer architectures, reinforcement learning, and reward modeling. Building on these, we will examine downstream issues such as alignment, jailbreaking (techniques to bypass model constraints), and red teaming (methods for identifying vulnerabilities).
Students will apply these concepts in guided, hands-on explorations of contemporary AI systems developed by frontier labs (e.g., Google, Meta, OpenAI, Anthropic). Following introductory exercises, student teams will design mixed-methods experiments to document and analyze model behavior in context. Exercises will unfold in open-source tools and environments such as Google Colab, computational notebooks, and remote scripting frameworks. Students will learn how to construct multi-shot prompts, programmatically interface with AI models, and assess system outputs using text-based metrics such as relevance, coherence, and specificity, as well as correlation analyses for bias detection. Alongside these technical investigations, the course integrates critical and humanities-based frameworks - including concepts such as technoslop, kitsch, and nostalgia - to sharpen the interpretive analysis of GenAI artifacts.
Spanning all of these activities we will seek to understand how the evolving landscape of AI systems is impacting trust, authorship, and knowledge production across disciplines.
Coding environments
Python, React.js, Virtual Computers and Jupyter Notebooks
Prerequisites
Curiosity, college level maths, at least one computer code creation class.
Syllabus
tba