Prediction Machines, Not Brains

Prediction

Part of: How AI Actually Works

Type "I'm running a little" into your phone and it suggests "late." A large language model works the same way: it is autocomplete that read most of the internet. Once that idea is clear, most of the rest of how AI behaves follows from it. What it is A language model is a program that does exactly one thing: given some text, it predicts the next chunk of text. That's it. There is no module for "reasoning," no separate "memory bank," no truth-checker. The entire system is a giant function that takes the text so far and outputs a probability for every chunk that could come next. The key operation is called next-token prediction . The model scores thousands of candidate continuations, picks a likely one, adds it to the input, and predicts again. That loop is the entire process. How it works Walk through one answer being built: 1. You send text. Your question becomes the starting input, called the context . 2. The model predicts. It assigns a likelihood to every possible next chunk and selects from the top candidates. 3. It appends. The chosen chunk is glued onto the input. 4. It repeats this predict-and-append cycle until it emits a special stop signal. There is no hidden step where th

Challenge: Greedy Decoder