Understand AI Better than 90% of People in 2 Minutes
Part 1: Tokens
When it comes to AI, everyone has an opinion about how it will affect the future of ____ (fill in the blank). But after discussing it with a lot of people over the past few months, I realized most people understand very little about how AI works.
In this series, I'll break down a few key concepts that will help you understand how AI works and help you avoid the confusion. These concepts will help you separate fact from fiction, which matters more than ever as people are bombarded with misinformation and fears about how AI will affect _____ (use same word from above).
Let's start with one of the most important basic concepts: tokens.
Tokens: What are they?
You may have heard about tokens and wondered how they fit into the AI big picture.
First, something surprising, AI doesn’t read words or even letters—it reads tokens. So, what are they?
Tokens are pieces of words. When text is entered, it's broken into smaller parts (tokens), like a few letters or even punctuation, and each token is processed separately.
For example, the sentence "ChatGPT is amazing!" might be broken into tokens like: "Chat", "G", "PT", " is", " amaz", "ing", "!".
Why is this important? Because tokens are used every time you interact with Claude, Gemini, ChatGPT, or any AI assistant—and when you use them all up, things can get… well, a little weird.
Have you ever noticed that after a long conversation, responses slow down and your chatbot starts to forget things? That’s because the more text you type (or paste), the more tokens are needed to process your request. At some point, you've used up all your tokens, and old used up tokens start to fall out of its memory.
Note: Try to avoid pasting large blocks of text, or uploading huge documents into the conversation, this will help conserve those valuable tokens.
Earlier AI models could only handle about 4,000 tokens at a time. Today, a paid subscription often supports more than 128K (tens of thousands of words), and the most recent versions of large models, like Claude and Gemini in advanced mode, can handle as many as 1 million tokens in a single context window.
If you understand just one thing about how AI works, tokens are the basic building units and are the foundation of how LLMs process information. Knowing what they are and how they're used puts you ahead of 90% of people when it comes to understanding how AI actually works.
Next up: how context windows work and why they are the real limit behind what AI can remember, see Part 2 for more about Context Windows.
Comments
Post a Comment