Conversation

Notices

1. Embed this notice
   Molly White (molly0xfff@hachyderm.io)'s status on Tuesday, 09-Jan-2024 08:15:24 JST Molly White

   on OpenAI's "our whole business model is yoinking copyrighted material", and some arguments i've been hearing recently from AI advocates about rolling back the scope of copyright law

   https://www.youtube.com/watch?v=sLYzYWOPJbk

   #OpenAI #AI #copyright

   • Embed this notice
     Evan Prodromou (evan@cosocial.ca)'s status on Tuesday, 09-Jan-2024 08:27:46 JST Evan Prodromou

     in reply to

     @molly0xfff This is a really good take! I have a question for you. Are you familiar with any of the collective rights schemes for copyrighted materials? It's a way to balance the rights of copyright holders with the needs of users to access that information at scale. These schemes have emerged several times in the 1800s and 1900s as new technologies arose. LLMs might be a similar case, and collective rights management might be an appropriate solution.

     https://en.wikipedia.org/wiki/Collective_rights_management

     clacke likes this.
   • Embed this notice
     Evan Prodromou (evan@cosocial.ca)'s status on Tuesday, 09-Jan-2024 10:45:57 JST Evan Prodromou

     in reply to

     • Luis Villa

     @luis_in_brief @molly0xfff I kind of figured they arose in times of crisis caused by technological changes. Especially when the new technology is wildly popular, like radio broadcasts or recorded music. Changing copyright law is a way to cut the Gordian knot.

   • Embed this notice
     Luis Villa (luis_in_brief@social.coop)'s status on Tuesday, 09-Jan-2024 10:45:58 JST Luis Villa

     in reply to

     • Evan Prodromou

     @evan @molly0xfff As with most monopolists, they tend to suck for everyone on both sides of the equation (artists and consumers) but they may still be a least-bad option. 🤷🏽

     (Note that they did not arise for books, websites, or (AFAIK) movies, so to what extent they arise organically as a result of technological change is questionable. But obviously if we’re going to see new ones in our lifetimes, now is probably the time.)

   • Embed this notice
     Evan Prodromou (evan@cosocial.ca)'s status on Tuesday, 09-Jan-2024 10:53:12 JST Evan Prodromou

     in reply to

     • Luis Villa

     @luis_in_brief @molly0xfff training an LLM on a corpus of text and then offering access to that LLM as a service isn't the same as copying one piece of text and redistributing it. But it's also not the same as reading a book and giving a speech about the subject of the book. It's something else, and probably the right way to handle the challenge is to adapt copyright to handle it.

   • Embed this notice
     Evan Prodromou (evan@cosocial.ca)'s status on Tuesday, 09-Jan-2024 10:56:24 JST Evan Prodromou

     in reply to

     • Luis Villa

     @luis_in_brief @molly0xfff one big problem with earlier technology changes was that they were wildly popular. Politicians had a lot of incentive to make up new structures to deescalate the conflict. ChatGPT is very popular, too, and I think there's an equal incentive to force a compromise rather than let the NYT force ChatGPT off the Internet.

   • Embed this notice
     glorp (glorp@infosec.exchange)'s status on Monday, 15-Jan-2024 16:22:12 JST glorp

     in reply to

     @molly0xfff the most egregious thing about this is that they yoink copyrighted material on one hand but then argue that others training on gpt4 output should be forbidden

     clacke likes this.