Knowledge cutoff
Also called training cutoff, cutoff date
The date a model's training data ends. It knows little or nothing about what happened after that.
A model's training data is collected up to a point, then training starts, then months of testing pass before release. So a brand-new model already lags the news, typically by a few months, and the gap grows every week it stays in use.
Without a web search, the model can't know about anything after its cutoff. Worse, it may not notice: it can describe last year's prices, rules or product names as if they were current. Apps often tell the model today's date, which helps it realise time has passed, but it still doesn't know what changed.
For example
Ask a model without web search about an election held after its cutoff, and it may name the favourite from before, sounding sure.
Put it to use
Where you'll meet knowledge cutoff in ChatGPT, Gemini and Claude, how to use it there, and the mix-up most people make.
Daily Bytes Plus is in the Android app.