Knowledge Cutoff
LLM & Language ModelsThe date after which an AI model has no training data — it simply doesn't know about events, products, or information that emerged after this date.
Every language model has a knowledge cutoff — the date when its training data stops. GPT-4's cutoff is April 2024, Claude's is early 2025. Ask about events after the cutoff and the model will either admit ignorance or (worse) hallucinate an answer.
Knowledge cutoffs are why AI tools integrate web search. ChatGPT's Browse mode, Perplexity's real-time search, and Claude's web access all exist to bridge the gap between static training data and current information.
The cutoff matters most for: recent events, new product releases, updated pricing, current stock prices, sports scores, and anything that changes frequently. For stable knowledge (how to write Python, what the French Revolution was), the cutoff rarely matters.
Real-World Example
If an AI doesn't know about a product launched last month that's probably its knowledge cutoff — not a flaw. Use tools with web search (like Perplexity) for current information.
Related Terms
More in LLM & Language Models
FAQ
What is Knowledge Cutoff?
The date after which an AI model has no training data — it simply doesn't know about events, products, or information that emerged after this date.
How is Knowledge Cutoff used in practice?
If an AI doesn't know about a product launched last month that's probably its knowledge cutoff — not a flaw. Use tools with web search (like Perplexity) for current information.
What concepts are related to Knowledge Cutoff?
Key related concepts include Hallucination, RAG (Retrieval-Augmented Generation), Training Data, Pre-training. Understanding these together gives a more complete picture of how Knowledge Cutoff fits into the AI landscape.