It's a case of mutually assured destruction: to charge someone with impersonation, they would have to admit that they saved data that was supposed to be only for age verification and then deleted.

One runs a mailbox app on any old disused android phone, it temporary stores content and deliver it to the main unit once the connection is restored. Bit simpler to install an app and scan a qr code for the average user compared to even configuring an XMPP client IMO.

Conversely, I have seen this opinion expressed a few times. I can’t judge the accuracy but there seem to be more than a few people sharing it.

Pretty confident that's the intention of that name

OpenAI has stated that its models were trained on publicly available and licensed data. There is no confirmed evidence that ChatGPT was specifically trained on copyrighted books like J.K. Rowling’s Harry Potter series. The company has not disclosed the full details of its training data.