Building AI Tools with Large Language Mo

In the past year, Large Language Models (LLMs) built using transformer frameworks have emerged as the fastest-growing area of research and investment in AI/machine learning. Recent releases of chatbots such as ChatGPT (OpenAI), Bing (Microsoft), and Bard (Google) quickly reached hundreds of millions of users and have become the face of artificial intelligence for consumers. There has also been an explosion in the number of applications that depend on LLMs for a variety of more specialized tasks. Recent models have shown impressive performance on both canonical machine learning tasks and for everyday use, yet are in many ways poorly understood and, in some cases, exhibit unexpected and potentially harmful behavior.

Policymakers, analysts, and non-profit and industry leaders need an understanding of these models to take advantage of the opportunities they present and to mitigate potential harms. This course provides an overview of Large Language Models and gives students hands-on experience with various ways of interacting with LLMs. Students will learn to interpret model evaluation metrics, and we will discuss safety and ethics in applied contexts. Prerequisite: Working Python knowledge OR Python for Public Policy (U6504) OR Intro to Text Analysis in Python (U6502).

This 7 week class will provide an introduction to LLMs, then outline best practices for interacting with these models for task completion both manually and programmatically (using Python). We will also discuss safety and alignment concerns in the context of task evaluation. Students will implement an LLM-based application that enables or streamlines a policy-relevant task or analysis, and demonstrate and evaluate its usage in a presentation and report.