Clicked Gallery

What is a knowledge cutoff?

Highlighted from a real engineering doc. Explained by Clicked.

Used in a sentence

Engineering Notes · AI Systems

The model's knowledge cutoff means nothing published after that date is stored in what it learned.

The reader highlighted one word in the docs. Clicked explained the technical term “knowledge cutoff” in simple terms:

Explained in three depths

Same facts, different vibe — Slang mode 😎

The Clicked way

●○○

Overview

A knowledge cutoff is the last date covered by the material a model was trained on. It limits one thing only: what the model can recall unaided. Hand it a search result, an uploaded file or text you paste, and it reads that perfectly well. Later fine-tuning can add knowledge too, so a fine-tuned model may know things the base model never saw.
●○○

Overview

A knowledge cutoff is the date the training material stops. Anything published after it never made it in, so the model cannot recall it unaided and will happily talk around the gap instead. Show it a fresh article and it reads that perfectly well, which is why every browsing feature exists. Fine-tuning can also push new facts in, so the date is a floor rather than a ceiling. 😎

A quick take — often all you need.

●●○

Detail

The material used to build a model gets collected up to a certain point and no further, and that boundary is the cutoff. Ask a model about an election held after it and there is nothing stored to draw on. The awkward part is that it does not experience the gap as a gap, so instead of stopping it may assemble a confident answer from what it did read. Why not just keep training it as events happen? Because training is one enormous run costing a fortune, so what a model knows is a photograph rather than a live feed. This is exactly why web search and retrieval-augmented generation exist. Both work by fetching current material and placing it in front of the model at the moment you ask, which sidesteps the cutoff entirely. A model with search enabled can discuss this morning's news. The same model without it is guessing from a year ago. The difference is what it was handed, not what it is. Models are also unreliable about their own cutoff date, since that date is just another fact somebody had to write into the training material.
●●○

Detail

Somebody gathered the training material up to a certain date and then stopped, and that date is the cutoff. Everything after it is missing from what the model can recall on its own. So ask about last week with search switched off and you may get something invented, delivered with total confidence. It is not lying. It has no sense that a gap exists, so it fills the space with whatever it does have. Switch search on and the same question works, because now the article is being handed over rather than remembered. Fine-tuning can push new facts in afterwards, though retrieval is the more dependable route. Models are also frequently wrong about when their own cutoff falls, which is a strange blind spot until you realise the date was just one more line somebody had to type in. 😎

Want more? One click digs deeper.

●●●

Analogy

Someone back from a year at sea. Whatever happened while they were gone, they were not there for any of it. Hand them today's paper and they will discuss it as well as anyone, because there is nothing wrong with their reasoning. The problem only appears when nobody hands them the paper and they answer anyway, from how things looked when they left.
Deepest level reached.
●●●

Analogy

Someone coming back from a ten day silent retreat. No phone, no news, nothing. They probably learned a great deal about themselves, but they have no idea that bell bottoms are cool again. Show them a headline and they are up to speed in seconds. The trouble only starts if nobody shows them anything and they answer anyway, from the world as it stood when they handed over their phone. 😎
Deepest level reached.

Unfamiliar concept? A real-world example makes it click — fresh analogies on tap.

AI explanations may contain errors · Not professional advice

Formal definition — The same term, explained the usual way

A knowledge cutoff is the latest date represented in the corpus used to train a model, beyond which no information is encoded in its parameters. It follows from training being a discrete process rather than a continuous one, and it constrains parametric knowledge only. Subsequent fine-tuning can extend that knowledge, though retrieval remains more reliable for factual currency. Information supplied at inference time, whether through retrieval, web search or a pasted document, is unaffected by the cutoff and can be used normally.

Want Clicked to explain terms like “knowledge cutoff” directly in your browser — including on PDFs?

Add to Chrome — Free

50 free Explanations · No credit card required