Why it cannot know your files

The weights were finished on a date, and your documents were never in them.

Part of the Memory and truth track on lAItest.

Ask a model about the contract you signed last Tuesday. It will answer. It will also be making the answer up.

Nothing in its weights has ever seen that file.

Training ends. The weights stop changing.

A trained model is a fixed pile of numbers. Text written after training finished cannot be inside it. Neither can text that was never published — your files, your database, your email. Those sit on a disk the model has no way to reach, because a model has no way to reach anything.

A common misconception

Commonly believed: The model is online, so when it does not know something it goes and looks it up.

Actually: By default nothing is connected. A model has its weights and the text you hand it, and that is the entire universe of a request. Search, files, databases: each one is a separate program that fetches text and pastes it into the prompt before the model runs. Building that plumbing well is the subject of this track.

Every set of weights has a finish line

In one sentence

A model knows two things: what was in its training text, and what you put in the prompt. There is no third source.