there are frameworks in place for this already, like GDPR. my comment is not about it being related in any way, it shows how you can see a certain decision making process, that is hostile to anything new
If openai perceives this to be just a problem for one 60M people country, not particularly known for tech enterprises, they may just shrug it off and close off service to Italy and call it a day.
That's exactly what they did. My plus account got refunded and blocked. VPN only allows me to access the free service, so no GPT-4 for me.
So unless other countries are following on and ensuring that tech giants are respecting privacy, this will just hurt Italian citizens and their ability to compete globally.
That's a political question, I just want to point out that this measure is backed by the GDPR, for better or worse.
At least some of the big companies take this seriously, because I've worked in consultants and seen it. It would be completely reasonable for OpenAI to be more private in the paid models - make the weights readonly, encrypt conversations, allow permanent deletion.
For the free models, I'd find it OK to train on the input, but GDPR says users should be able to request data deletion - not sure how viable that is for LLMs, where training frons scratch regularly (without the requested deletions) is probably expensive as fuck.
The internet has been a wild west with lots of privacy abuse, it's good to see some rules and enforcement!
It was expensive but we had to retrain models frequently anyway because of other reasons so this just added a more strict cadence and time bound checks but didn't really inflate costs dramatically