I’ve been having trouble generating coding interview questions in a consistent manner. This is the output I am looking for:
GPT-3 is able to generate this, but not in a consistent manner. I was hoping to automate this process with the OpenAI Python Library, so I was wondering if there was a way to restructure the prompt itself to help guide GPT-3 better.
If you provide it more examples and ask it to produce 3 instead of 5 questions it will likely generate good questions more consistently. Also, check your settings on the side - the greater the temperature, the more unique the answers will be. You don’t need to specify “neatly formatted” either if you provide it an example or two or that it is a “prompt”.
So it could look like:
This is an interview question generator.
Generate a list of 3 questions for an interview with a beginner Java developer.
then set your start text to “1.” and your stop text to “4.”