Lesson 1: What is AI?
Text Prediction vs LLMs
Text prediction on your phone and LLMs are similar in that they are trained on text data (words and sentences) and use this to predict the next words in a sentence. However, there are differences in how the training is done, the amount of training data used, and what their capabilities are.
The scale of the computers used to train and power LLMs is massive compared to text prediction models. Often, these computers are in data centres, which are large buildings full of computers that are used to store and process data. You can click through the images below to see some photos taken inside and outside of data centres.
The table below summarises the key differences between text prediction and LLMs.
| Text prediction on your phone | LLMs |
|---|---|
| Tiny model (a few megabytes). Looks at only 1-3 previous words. Optimised for speed and accuracy. Rarely invents new ideas. | Massive models built using deep learning (lots of neural networks). Understand full conversations and long context. |
| Looks at: 1-3 words | Looks at: many paragraphs or full conversations |
| Model size: a few megabytes | Model size: many gigabytes (stored on server computers, usually in data centres) |
| Runs locally on your phone; no internet needed | Runs on powerful server computers |
| Predicts the next word based on how common they appear in training data | Generates new ideas, explanations, reasoning, and creative text |
You'll learn much more about how LLMs work in future lessons in this course.