A large language model () is a deep-learning model trained on large amounts of text to process and generate natural language. When you a conversational AI service, a model of this kind writes the response.
How a model generates a response
The model does not retrieve a finished answer from a database. It breaks your text into and estimates which token is likely to continue the text so far. It adds the selected token to the response and repeats the process until the answer is complete. During training, the model learned how words and expressions commonly relate to one another.
That is why a model can write fluently about almost any subject. This process often produces accurate text, but it can also produce errors.
A model and a service are different things
The AI service you use is a complete product: a language model together with the features and settings built around it. Services built around the same model can therefore behave differently.
Fluent does not mean checked
Fluency tells you that the model produced likely text; it does not tell you the information is correct. Check an important claim against its original source before using it.