Author: Barry Zhang

Contributors: Nahom Yimam, Daniel Chang

Last Updated: 05/05/2023

Goal of this article:

I am neck-deep in the rapid tides of LLM these days. I find these things incredibly beautiful and want to share my perspective. With the help of AI, it is easy to learn just about anything, except for the taxonomy of the unknown. So, this article aims to provide a very basic and incomplete taxonomy that you can hopefully expand your own knowledge from.

Sections:

• A Cool Chart
• The Components
• How they connect
• The AI Tech Stack
• What is not written here

A Cool Chart:

This entire article will revolve around the chart that I made for the basic data flow of LLMs. I will dive into each of the components and explain how they connect.

What’s NOT shown in the diagram (and the rest of the article): model architecture, AI infrastructure, implementation details, ANYTHING agentic (which will be covered in another article), etc.

Components:

User:

• In this article, we will refer to “User” as the entity that provides inputs to the system (e.g. chatting with chatGPT) as well as implicit/explicit feedback to the system (e.g. thumbs up/comments/session dropoff)