Setting your own data cutoff

Hi all,

For my research, I want to use GPT4 and set my own data cutoff date. For example, set it such that the model only uses data up until 2019. Does anyone know if there is a way to do this?

Thanks in advance.



The information that the AI has been pretrained on is not delineated by date. It is simply as a whole used to predict text. “A cute furry” → “kitten” token prediction for example, or “the royal leader of the UK is”, is not informed by more than the accessory knowledge the AI may be able to infer.

You can give the AI a system message about its knowledge cutoff date. This will only give you denials that don’t include the new date, or a pathway into new hallucinations about events the AI doesn’t know.

Roleplay as an AI that was built in 2010 and doesn’t know anything beyond that date” is direct but also not successful. “as an AI from 2010, I don’t know about the 2016 US presidential election, but it was in fact won by Donald Trump” a typical style of answer.

1 Like