22 ms·
This seems wrong on the face of it. Even assuming open source AI is everything the article promises, surely taking an open source model then adding a pile of sp
by flir 3y ago
This seems wrong on the face of it. Even assuming open source AI is everything the article promises, surely taking an open source model then adding a pile of specialist private data and lots of training compute is going to make something just a bit better?
I guess the closest software analogy is AWS. Leverage open source hard, but don't bother about giving back.
- toenail 3y agoI guess AWS never gave OpenSearch back to the open source community? Or anything on https://aws.amazon.com/opensource/ https://aws.amazon.com/opensource/
- flir 3y agoWhen I was writing that comment, I did come across this article from Apr 2023: https://www.infoworld.com/article/3694090/amazon-s-quiet-open-source-revolution.html https://www.infoworld.com/article/3694090/amazon-s-quiet-ope... But please don't worry too much about the example I used, it's just a tangent. What's wrong with my argument about closed-source AI always having an edge? (Actually, I can think of one possibility - when a company has an incentive to open-source AI in order to commoditize their complement. Content aggregation companies might want to commoditize content-generating AI in the same way that hardware companies wanted to commoditize software. Same argument holds for Nvidia maybe - what better way to sell silicon than to make sure there are a lot of really useful open source models out there to take advantage of it?)