Knowledge Cutoff

Knowledge cutoff is the point in time where a model's training data stops — in other words, the last date the model actually knows about the world. If a model's cutoff is this past January and you ask it about last week's news, it either doesn't know, or worse, makes something up (hallucinates) as if it does.

This is exactly why chatbots add search features. A model's knowledge is frozen at a certain point, and search is how it "reheats" that knowledge with current information. So there's no need to be surprised when a chatbot doesn't know about yesterday's announcement — and it's also why you should check whether search is turned on when asking about anything time-sensitive.

1. Ask a chatbot, "What's your knowledge cutoff?"
2. Turn off search and ask about something that happened after that date (a recent sports result, for example) — it won't know, or will dodge the question.
3. Turn search on and ask the same question — you'll get an up-to-date answer. This is a quick way to feel the difference between "what the model remembers" and "what search reheats for it."