LLMs Under the Hood

How does an LLM work? What happens between sending a prompt to an LLM and receiving a response? This course explores the architecture of an LLM, each subcomponent that runs, and how LLM’s are built and trained. The seminar does not assume any mathematical or coding background but will go into technical detail on what LLM’s are doing behind the scenes.

ABOUT THE SPEAKER
Quinn Theobald is a software developer and writer. They studied Information Science at Cornell University and worked as a software engineer at Epic Systems, where they developed projects deploying LLMs in the health tech field. They've taught classes on data science and LLMs at Cornell and at Epic Systems.
Email: [quinntheobald@gmail.com](mailto:quinntheobald@gmail.com)
LinkedIn: https://www.linkedin.com/in/quinn-theobald-aa496220a/

Quinn Theobald will pull from their experience working on high-impact LLM projects in the medical tech field to discuss how LLMs operate. The seminar will focus on the architecture of the GPT-2 model because of its simplicity, but the concepts are broadly applicable for other LLMs. The seminar will cover the following topics:

- Neural Networks: how they are built and how they work.
- “Transformer” models that convert input text into context-rich numeric embeddings. This is how an LLM “understands” prompts.
- Attention layers that evaluate relationships between words. This was the last crucial breakthrough that made LLM development possible in 2017.
- Training and how LLMs are constructed by the major AI builders.
- A discussion on bias in AI.
The seminar will have openings for questions and discussion. The seminar will last two hours.