Many companies are exploring what can be done with LLMs, and LLM skills are in high demand. However, large language models are a huge field of study that is growing daily. Knowing where to start isn’t easy, and best practices are changing rapidly.

While there are a billion LLM tutorials, they often show how to build a toy demo without explaining the underlying foundational principles. These are key to understanding the behaviors of LLMs and deploying robust LLM-powered applications.

This workshop seeks to address this gap. Each 1-hour session is self-contained, i.e. fresh code is provided at the start of each session, so attendees can drop in and drop out without missing anything.

Session 1: How does an LLM work?
* Overview of the training process
* Overview of the generation (output) process
* What is "truth" to an LLM?
* The difference between creativity and hallucination

Session 2: Build a chatbot
* What happens when an LLM doesn’t know the answer?
* The difference between an LLM and an LLM-powered product application
* How can we introduce new knowledge to the LLM?
* How do we test & evaluate the LLM’s output?
* LLM caching to save on cost

Session 3: Build a knowledge base question-answering tool
* Explore model fine-tuning vs Retrieval-Augmented Generation (RAG)
* Overview of RAG & an implementation
* Data provenance concerns & approaches
* IP concerns & approaches

Session 4: Prompt engineering
* Techniques & approaches
* Chain of Thought
* Few- and many-shot learning
* Prompt injection

Notes to organizer/selection committee: each 1-hour session is self-contained, i.e. fresh code is provided at the start of each session, so attendees can drop in and drop out without missing anything.