Glossary
A large language model is a system trained on very large amounts of text to predict likely continuations of language. That single capability is what lets it summarise, answer, translate, classify and draft.
A model does not look up an answer in a database; it produces text that is statistically plausible given everything it has seen and the prompt in front of it. Understanding that one fact explains most of the behaviour that surprises people.
The same process that produces a well-formed sentence produces a well-formed sentence that is wrong, with no internal signal distinguishing the two. Confidence in the output is a property of the language, not of the facts.
A model given relevant source material at the time of the question performs very differently from one asked to answer from training alone. This is why serious business applications retrieve context first rather than trusting recall.
It is 'what does it have access to, what happens when it is unsure, and who is accountable for the answer'. Those are design decisions, not model choices.
Because they generate plausible text rather than retrieving verified facts. When the training data is thin or the prompt is ambiguous, the most plausible continuation can be false, and the model has no separate mechanism telling it so.
Almost never. For most business problems, a hosted model given access to your own documents at question time performs better than a fine-tuned one, at a fraction of the cost and maintenance burden.
A large language model is a system trained on very large amounts of text to predict likely continuations of language. That single capability is what lets it summarise, answer, translate, classify and draft.
