A large language model, or LLM, is a neural network trained on enormous amounts of text. Given some text, it predicts what should come next, one small piece at a time.
That single ability turns out to cover a lot. Answering a question, writing code, translating, summarising: all of them are just predicting what text follows. The model is not looking anything up. It is continuing a pattern.
Think of it like this. Think of someone who has read most of the internet but remembers none of it exactly. Ask them anything and they answer from a very strong sense of how sentences like that usually go. Often right, occasionally confident and wrong.
How it works
- Text becomes tokens. Your input is split into small pieces, roughly three quarters of a word each.
- The model predicts the next token. It produces a probability for every token it knows, then picks one.
- It repeats. Each new token is added to the input and the process runs again, which is why answers appear word by word.
- Parameters are what it learned. These are the numbers adjusted during training. More is not automatically better.
Common mistakes
- "It understands what it is saying." It predicts statistically. That produces useful answers without producing understanding.
- "It looks facts up." There is no database inside. Facts are approximated from training, which is why it can invent them.
- "Bigger is always smarter." Training data quality, efficiency and alignment matter as much as parameter count.
- "An LLM is a chatbot." A chatbot is one application. The model is a general text function you can build many things on.
Trade-offs
- Flexible, not reliable. It handles vague, messy language that normal code cannot, and gives no guarantee of correctness.
- Bigger costs more. More capable models are slower and more expensive per request.
- Knowledge has a cutoff. Anything after training is unknown unless you supply it in the prompt.
