That they’d focus on file sharing over transformation or outputs is exactly the risk I warned the companies about in my AI report. Most datasets, like RefinedWe

That they’d focus on file sharing over transformation or outputs is exactly the risk I warned the companies about in my AI report. Most datasets, like RefinedWeb and The Pile, also require sharing copyrighted workers between people who are not licensed to do that. Many works also prohibit commercial use or have patents on them. They need to make datasets which don’t have this problem or have entities in Singapore train the foundation models within their rules. The latter has a TDM exemption that would let AI’s use much of the Internet, maybe GPL code, licensed/purchased works they digitize, etc. Very flexible.