Knowledge cutoff

Everything the model has baked in was collected up to a date — the industry calls it the knowledge cutoff — and coverage thins on the approach: training does not sample the world's final months as densely as the years before them, so the built-in world does not end at a wall so much as fade toward one. Treat the published date as a warning line, not a guarantee on either side of it.

Anything newer — the price, the version, the ruling, the person's new title — can only reach the model through the second door, the context window: pasted by you, or fetched in by a search the app ran. That is why the window carries a supplied date block: the model does not know today's date until the app pastes it in.

Two moves fall straight out. For anything that could have changed, either watch the search step run — the receipt — or paste the current fact yourself; if neither happened, the answer was drawn from the fading world behind the warning line. And never ask the model for its own cutoff: that produces a completion, right only when the real date was pasted in or deliberately trained in, and you cannot tell which from your chair. Each model's cutoff is a lookup in its product's documentation. The documentation is the checkable source.

Say it back in one line: why is "what is your knowledge cutoff?" the wrong question to ask the model itself?