BeginnerLesson 3 of 69 min read
LLMs: the word-predicting machine
A course from the Academy: AI finally made clear: the foundations
Go to the quiz ↓ChatGPT, Claude, Gemini, DeepSeek… they all belong to the same family: LLMs (Large Language Models). Their secret is disconcertingly simple.
A single talent: predicting what comes next
An LLM has been trained on a single task: guessing the next word. You give it "The sky is" and it has to suggest "blue". Trillions of times, on everything humanity has ever written. To become truly good at this game, the model has been forced to absorb grammar, facts, logic, and styles, because you can't properly complete a sentence about law or physics without having picked up something about those fields.
When you chat with an AI, that's all it does: predict the next word, then the next, then the next… The entire conversation is a chain of predictions.
Tokens, the currency of language
The model doesn't see words but tokens: fragments of text (a short word, a chunk of a long word, punctuation). "Anticonstitutionnellement" will be split into several tokens; "le" into just one. Everything is measured in tokens: the length of your question, the length of the answer, the price of AI services, and the size of the model's working memory.
Why it seems intelligent
Because by endlessly predicting human text, the model remarkably mimics the human reasoning that produced that text. It's an open debate: is imitating reasoning the same as reasoning? Either way, remember that the LLM doesn't "know" it's answering you; it completes a text, brilliantly.
Key concepts
Complete the lesson
A 4-question quiz: 3 correct answers are enough, plus an exercise to practise. It opens in the Academy, on this same lesson.
Take the quiz and complete the lesson Free. With an account, your progress is saved and leads to the course certificate.