Can you fine tune by training with web data?

I see that gpt-3.5 has an Alpha for reading a webpage (or series of web pages) and then responding about them.

Is it possible to fine tune a model via this method as opposed to fine tuning with high quality example prompts and completions?

You can collect your own data and save it in a vector database and then use chatGPT to ask questions about your data.

You may find it helpful to take a look at this

I also created an example in Nodejs, that will read websites and give back the results to ChatGPT