The knowledge cutoff problem

Models know a lot, but not your data

A model is trained on a snapshot. It has never seen your internal documents, your prices, or anything published after training, and it cannot tell you which of those categories a question falls into.

Asked about something it does not know, a model produces a plausible answer rather than an empty one. Retrieval fixes this by putting the actual facts in front of it at request time.

Fine-tuning teaches style and format far more reliably than facts. Changing knowledge means retraining; changing a retrieved document means saving a file.