Training data with names


I have a collection of over 100 k questions and answers. The collection contains names in the questions and answers that will not be relevant for the new questions the fine tuned model will answer. But the questions asked will certainly contain names as well.

Here is an example of a prompt and completion.

Prompt: One of our users Karen Baldwin has problem when she tries to login. Maybe her password is wrong? Can i get assistant?

Completion: Tell Karen that she can get a new password from the login page e-mailed to her e-mailadress. Just click on the link Forgot password. 

Steven, Technical Support agent

There are to many posts to do this manualy so I will have to do this automated. But what do I replace the names with?

Can you use some kind of variables for the names instead like [name]?

If someone asks a similair question I dont want the model to answer with a name that has nothing to do with the question.

For example if someone asks: How do I get a new password. The model can not answer: karen can change her password from the login page Forgott password function.