Next-word prediction
Also called next-token prediction
How a language model writes: it keeps picking a likely next piece of text, given everything before it.
Underneath every reply is one repeated step. The model looks at everything so far, works out how likely each possible next token is, picks one, adds it, and does it again, hundreds of times.
That simple step, trained at enormous scale and then tuned to follow instructions, is enough to write code and explain tax rules. It also explains the famous failure. The model is built to produce text that fits, and a confident, specific-sounding answer fits a question very well, whether or not it's true. Nothing in the step itself checks facts; that has to come from sources, tools or you.
For example
After "The capital of France is", "Paris" is overwhelmingly likely. After "The founder of my local bakery is", the model still produces a likely-sounding name.
Put it to use
Where you'll meet next-word prediction in ChatGPT, Gemini and Claude, how to use it there, and the mix-up most people make.
Daily Bytes Plus is in the Android app.