Conversation

Notices

1. Embed this notice
   Christopher Hudson (chrishudsonjr@mastodon.social)'s status on Wednesday, 10-Jul-2024 03:48:13 JST Christopher Hudson

   • Aaron Ross Powell ☸️

   So much criticism around #LLMs, #AIart, and #artificialintelligence more generally is so clearly flawed. @arossp does a great job in this piece illustrating why.

   https://www.reimaginingliberty.com/p/ai-theft-thought-experiment
   #tech #intellectualproperty #art #copyright

   • BeAware :fediverse: repeated this.
   • Embed this notice
     BeAware :fediverse: (beaware@social.beaware.live)'s status on Wednesday, 10-Jul-2024 03:48:13 JST BeAware :fediverse:

     in reply to

     • Aaron Ross Powell ☸️

     @chrishudsonjr @arossp I feel the article might be slightly flawed because these things are pretty easy to train these days and anyone with some technology knowledge can train on any data they wish. So, in theory (haven't specifically heard of anyone doing this but it can be done) someone can just train it on whatever copyrighted media they wanted and use it just as if the companies themselves trained it.

     The biggest difference is, when you train these things on VAST amounts of data like these companies have, they can definitely be incorrect. However, when there's a specific, small amount of data that it's trained on, I'd be willing to bet that the outcomes are more accurate.

     Mr. Joe Schmo from Wyoming, won't ask for permissions from anyone before doing what they wish. Then it's just as easily uploaded to the internet.

     The cat is out of the bag. There's no way to bring it back in.