3 ms·
> GPT-3 was even openly trained on books from a literal pirate torrent tracker What's stopping a human from reading all of the same books?
by dorkwood 3y ago
> GPT-3 was even openly trained on books from a literal pirate torrent tracker
What's stopping a human from reading all of the same books?
- rectang 3y agoAs the previous commenter indicated, scale — it’s not physically possible for a human to read that many books. Even the most dedicated readers will only read a few thousand books in their lifetime. https://www.mentalfloss.com/article/93594/heres-how-many-books-you-can-expect-read-you-die https://www.mentalfloss.com/article/93594/heres-how-many-boo... ChatGPT is trained on orders of magnitude more books than a human can consume. https://torrentfreak.com/authors-accuse-openai-of-using-pirate-sites-to-train-chatgpt-230630/ https://torrentfreak.com/authors-accuse-openai-of-using-pira... > OpenAI hasn’t disclosed the datasets that ChatGPT is trained on, but in an older paper two databases are referenced; “Books1” and “Books2”. The first one contains roughly 63,000 titles and the latter around 294,000 titles