Tools like ChatGPT are built on large language models, or LLMs. The core idea is simpler than it seems.
Direct answer
A large language model is trained on huge amounts of text to predict what comes next. It breaks text into small pieces called tokens, learns patterns between them, and when you ask something it generates a reply one token at a time. It doesn't look up facts, so it can sound confident while being wrong.
Step by step
- Text becomes tokens. Words are split into small pieces.
- Training. The model reads vast amounts of text and learns to predict the next token. Its internal numbers are adjusted until predictions are good.
- Fine-tuning. Extra training with examples and human feedback makes it more helpful and safer.
- Using it. Your question is turned into tokens. The model predicts the next token, adds it and repeats, producing the answer.
What it's good at
- Writing, rewriting and summarising
- Explaining ideas
- Translating
- Helping with code
Where it struggles
- Facts it hasn't seen or that are recent
- Exact maths and counting
- Knowing when it's wrong
- Long chains of reasoning without checks
Why it makes things up
It predicts plausible text. If it doesn't know, it may still produce a plausible answer. Check anything important. See how to use ChatGPT.
Beyond the model
An app can give a model tools (search, code, files), memory and rules. That is how assistants and AI agents are built. For the wider picture, see how AI works in simple terms.
Where Teyro fits
Teyro's The AI Stack course covers chips, models, data, agents, apps and evaluation, and you build your own AI assistant in your browser, one layer at a time. It is free to start.
The bottom line
An LLM is a very good text predictor. Useful, but verify.
