Surprise EU rollback of 'GDPR' digital-rights rules prompts alarm

https://lemmy.zip/post/52834246

Surprise EU rollback of 'GDPR' digital-rights rules prompts alarm - Lemmy.zip

cross-posted from: https://lemmy.zip/post/52834195 [https://lemmy.zip/post/52834195] > https://archive.is/je5sj [https://archive.is/je5sj] > > > “If adopted, these amendments would not simplify compliance but hollow out the GDPR’s and ePrivacy’s core guarantees: purpose limitation, accountability, and independent oversight,” Itxaso Dominguez de Olazabal, from the European Digital Rights group, told EUobserver. > > > The draft includes adjustments to what is considered “personal data,” a key component of the GDPR and protected by Article 8 of the Charter of Fundamental Rights of the European Union.

Doesn't seem terribly surprising to me, the existing rules make it very hard to make use of data for AI training in the EU. Other parts of the world have looser restrictions and they're developing AI like gangbusters as a result. The EU needed to either loosen up too or accept this entire sector of information tech being foreign-controlled, which would have its own major privacy and security problems.
There is nothing stopping the EU from going the DeepSeek route and just stealing the finished LLM’s from American companies. But the truth is that the EU shouldn’t want to have all these data centers training generative models. The us is already dedicating 4% of our electricity production to them, with people in states along the Great Lakes and Eastern seaboard seeing massive increases in their electric bills to pay for them (~30% for me in Ohio, ~75% for my brother in Virginia). I can understand if you are a technocratic neoliberal in the EU parliament that is taking bribes from tech firms why you would want this, but for anyone paying attention, rhe promises tech companies are making to burn hundreds billions of euros while gutting privacy, 🔏IP, and consumer protections at the top of the bubble makes no sense.

Deepseek was trained from scratch.

That aside, you're basically describing the second option I presented; letting everyone else do the AI thing instead.

DeepSeek is it’s own model, designed and trained from ground up. It’s a novel architecture even. Impressive work.

It’s not a ‘stolen from the US’ model.