This is a sample application that shows how you can use GPT to generate demo data.

There are two input files, sys_prompt.txt and user_prompt.txt that are feed into the GPT chat API. sys_prompt.txt contains generic instructions for how to generate demo data, while user_prompt.txt contains the specification of the data to generate.

The specification can be a CSV dump of a few rows of pre-existing data. Even if that data is not high quality, GPT can use that to generate better looking, more complete data. You might also get away with a more complicated format (if you want to provide hints and restrictions on the data to be generated); that might require updating the system prompt to describe your format.

I've provided a sample user_prompt.txt file to give you an idea of what an input might look like.

The code is set to generate 10 sample rows, but you can easily change that.

To generate 10 sample rows costs about 6 cents with GPT-4. GPT-3.5 tends to have problems in its generation: thus I prefer GPT-4, but perhaps more prompt work could fix those. It doesn't seem worthwhile, though, given how cheap GPT-4 is, unless you need high volumes of data. The code prints the tokens consumed both on the input (prompt) and output so that you can see how much it's costing to generate your demo data (using the current pricing from OpenAI).

Finally, since most LLMs offer very similar APIs to OpenAI, you could switch this to using some other vendor's API with minimal effort.