Re:Creators of AI-generated George Carlin special sued by la...
by dorkwood on Saturday January 27, 2024 @10:21PM
(#39162238)
> 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?
Re:Creators of AI-generated George Carlin special sued by la...
by rectang on Saturday January 27, 2024 @10:53PM
(#39162399)
As 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-boo...
ChatGPT is trained on orders of magnitude more books than a human can consume.
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
- The qualitative differences. You can't run an unbounded # of simultaneous and 24/7 instances of a natural human intelligence.
- Machine learning models are not people and do not have rights.
- Obtaining the training material itself requires massive copyright infringement. GPT-3 was even openly trained on books from a literal pirate torrent tracker.