Knowledge Cutoff
The Encyclopaedia That Was Printed in January
4 min read
The Encyclopaedia That Was Printed in January
Imagine an encyclopaedia printed in January 2024 — brilliant on everything before that date, silent on everything after.
If you ask it about an election that happened in March 2024, it has nothing. That's not a flaw — it's a limitation of when it was printed. AI models have the same limitation. Their training data has a cutoff date, after which they have no knowledge. Anything after that date is invisible to them unless you tell them.
In Plain English
An AI model's knowledge cutoff is the date after which it has no training data. It knows nothing about events, products, or changes that happened after that date unless you provide that information in the conversation.
The Technical Picture
The training corpus has a data collection cutoff date. Because model training takes months after data collection ends, and deployment follows training, there is typically a 6–12 month gap between the knowledge cutoff and when users interact with the model.
Real-World Examples
- Claude's knowledge cutoff is early 2025 — it may not know about very recent news
- Perplexity solves this by searching the web in real-time
- ChatGPT now includes web search to supplement its knowledge cutoff
Ask an AI about recent events — if they happened after its cutoff, it either won't know or will guess wrong.