4 ms·
I keep thinking that copyright lasts too long. It seems to me like any regulation around AI training is most likely to benefit a small handful of giant corporat
by TheAceOfHearts 2y ago
I keep thinking that copyright lasts too long. It seems to me like any regulation around AI training is most likely to benefit a small handful of giant corporations, which will be the only ones that can afford to license content and train newer models. And the small handful of mega corporations that own the rights to most modern media will get their payday. In the end culture ends up being owned and controlled by a few big corporate interests.
I'm sympathetic to artists that want to exclude their works from giant training datasets that primarily end up benefiting the big players in AI without giving anything back. In a way, the first big digital data heist on mankind was executed by social media. Founders of various social media sites became extremely rich thanks to regular people's content being posted and shared (often illegally), without giving anything back to creators.
- wslh 2y agoSeems like the "Cobra Effect" all over again [1]. [1] https://en.wikipedia.org/wiki/Perverse_incentive https://en.wikipedia.org/wiki/Perverse_incentive
- k__ 2y agoWe need one that prevents companies like Disney.
- Rinzler89 2y agoThis. And Google, and Microsoft, and Meta, etc.
- llm_trw 2y agoWe've already reached peak data and high quality datasets provide phenomal performance improvements. The tech itself will push for gatekeepers that create the mountains of training data required.