It's interesting that generative AI seems to have such tiny moats compared to other AI domains which have larger moats, mostly because they require private acce

It's interesting that generative AI seems to have such tiny moats compared to other AI domains which have larger moats, mostly because they require private access to data. It seems that "downloading the textual internet" isn't that hard after all, and the field is not evolved yet to ponder about legality to use scraped texts in training which might provide a moat. That would suddenly make the archives of big copyright owners like random house way more valuable...