But also worried that OpenAI has access to Lemmy posts.

They have always had access to Lemmy posts.

They don't, that's actually my implementation, I had to add custom functions so it can retrieve live data.

If you're interested in the technical implementation: I provide list of functions that I implemented to the GPT model.

When you ask for example for your most upvoted comment, it sees that I provided a function to get your most upvoted comment, so it tells me "call the function to get most upvoted comments with these parameters and tell me the result".

Afterwards the code calls the function, which gets the most upvoted comment and provides it back to the model and GPT then creates an answer using the result.

Pretty much anything can be implemented like that, that's how it can have access to live data.