Hacker News Reader

Comment by ronsor | Hacker News Reader

I'm not sure how public posts count as "private data".

For instance, under GDPR, you have the "right to be forgotten", which means your data (and any traces of it) must be deleted. If an ML model learns from it, perhaps (IANAL) that implies the model must be re-trained when a GDPR deletion request is processed? And it's assumed they are not doing this (because of course they're not)?

Just because the data was once public doesn't mean you are allowed to embed it in a model forever.

noyb : https://noyb.eu/en/twitters-ai-plans-hit-9-more-gdpr-complai...

There you'll find the arguments.

Public posts are by default copyrighted to the author of the post, unless the author grants an explicit license. LLM training is copying. Next question please.

"Public posts are by default copyrighted to the author of the post"

US Centric and also moot given you agree to twitters terms of service before your first post.

Copyright works more or less that way almost everywhere. European courts have tended to be sceptical of the idea that agreement to an EULA that no-one has ever read qualifies as consent, particularly in cases where there’s something unexpected lurking there.

Isn't this the case for all countries that accepted the Berne Convention?

>LLM training is copying. Next question please.

Says who? At best it's murky since AI companies argue that it's fair use, and that's entirely unfounded, eg. https://en.wikipedia.org/wiki/Authors_Guild,_Inc._v._Google,....

No it isn't. Better responses please.

Reply on news.ycombinator.com