Programming Generative AI Applications

Generative Artificial Intelligence is a type of AI that learns patterns from data to create new content in various types of media (text, images, audio, video). At its heart a generative AI system has a large language model (LLM) that is essentially a large (trillions of parameters) neural network that has been trained on a mix of vast amounts of data as well as human input. Applying generative AI to actual problems in business often requires that the LLM underlying the AI be customized to the business problem, either by attaching a data source (e.g., operating procedures, 10k reports, marketing plans, balance sheets, etc.) to the LLM (a process known as Retrieval-Augmented Generation or RAG) or by retraining the neural net with additional data (a process known as fine tuning). adjusting the parameters of the underlying LLM. Embedding generative AI into organizational processes requires
that we gather appropriate data and reprogram the LLM to use the data either through RAG or fine tuning.

The focus of this course is to give you a working knowledge of what it takes to customize and assemble a customized generative AI application. We will use OpenAI’s GPT as our base model and learn how to build a RAG and how to customize using simple fine tuning. About 50% of the class time will be devoted to a group project where you will, in small groups, build your own customized AI application. All programming will be in Python and we will use libraries like tensorflow, langchain and faiss.

STUDENTS WILL NEED TO COMPLETE AN INTRODUCTORY PYTHON CLASS (https://courseworks2.columbia.edu/courses/152704) OR PASS THE BASIC PYTHON QUALIFICATION EXAM (https://cbs-python.com/) BEFORE THE FIRST DAY OF CLASS. SEE https://academics.gsb.columbia.edu/python FOR DETAILS