Problem getting the same results from a request done with postman and from playground

Dear community,
i tested the GP3 Models to help me extract Name, Email and Phone Numbers from a given text.
It’s crazy how well the model can handle the task. I do run into a problem where the model makes up random emails or phone numbers and sent them back as a “valid” result. Some of the numbers look fake like 01234567 but others look like real numbers that are false.

Is there a way to leverage the power of the model but have it not make stuff up? I tried like 20 different prompt iterations and did run a lof of experiments but without luck.

Maybe someone has an idea of how to do it.
Help would be greatly appreciated.


Hi @patribreitenmoser

You can try to fine-tune. This should help you.

Fine-Tuning: OpenAI API