It predicts the next word
A language model has read an enormous amount of text and learned which words usually follow which. When you ask something, it builds the answer piece by piece, always choosing a likely continuation. These pieces are called tokens — parts of words.
Fluent is not the same as true
The model is trained to produce text that sounds right. It has no built-in fact checker, so it can state a wrong date, law or quote just as confidently as a correct one. This is called a hallucination.
Its knowledge has a date
A model learns from texts up to a certain point in time. Without web search it does not know newer prices, laws or events. Many assistants can now search the web — then check which sources they used.
Where it shines
- Drafts, rewording and summaries
- Explaining concepts in simple words
- Brainstorming and structuring ideas
- Translating and correcting language
- Code, formulas and spreadsheets — with checking
Why Slovenian costs more
Models split Slovenian words into more tokens than English ones, because they saw less Slovenian text. That means slightly slower answers, higher API costs and sometimes clumsier grammar.