AI
Knowledge Cutoff
A knowledge cutoff is the date up to which a language model's training data was collected. The model has no inherent awareness of anything that happened after that date unless the information is provided to it directly in a prompt or retrieved from an external, up-to-date source.
A model doesn't "know" its own knowledge cutoff in any deep sense — it can state a cutoff date if that fact was included in its training or system prompt, but it cannot reliably distinguish between things it knows accurately and things it's guessing at from patterns in older data, which is a common source of confidently wrong answers about recent events.
Retrieval-augmented generation and web-search tool integrations exist specifically to work around this limitation: rather than relying on the model's frozen training knowledge, the application fetches current information at query time and provides it as context, which the model then reasons over.