3 ms·
"Content you're allowed to scrape from the internet" is MUCH smaller than what LLMs have actually scraped, but they don't care about copyright. > The fears tha
by nicolas_17 3y ago
"Content you're allowed to scrape from the internet" is MUCH smaller than what LLMs have actually scraped, but they don't care about copyright.
> The fears that we have “used all the data” are incredibly unfounded.
The problem isn't whether we used all the real data or not, the problem is that it becomes increasingly difficult to distinguish real data from previous LLM outputs.
- Dylan16807 3y ago> "Content you're allowed to scrape from the internet" is MUCH smaller than what LLMs have actually scraped, but they don't care about copyright. I don't know about that. If you scraped the same data and ran a search engine I think people would generally say you're fine. The copyright issue isn't the scraping step.