If you can't determine whether this data exists on the web, you need to be careful when using LLMs. On the other hand, if you have a lot of experience with edge cases where the web is absent, I think it can be very advantageous.