Knowledge cutoff
Also: Wissensstand · Trainingsstichtag
The point in time up to which a model’s training data extends.
A model does not know what happened after its knowledge cutoff from its own knowledge. Asked about current events, it may give outdated or invented answers.
Many assistants compensate with web search or RAG: they fetch current information at runtime instead of relying solely on what they learned in training.
General information, not legal or professional advice.
