Conversation

Notices

1. Embed this notice
   Nathan Schneider (ntnsndr@social.coop)'s status on Wednesday, 11-Sep-2024 06:52:03 JST Nathan Schneider

   Just got some useful debugging (doesn't really feel like editing) on a draft paper from an AI running slowly and locally on my machine. No data sent to the cloud. This feels better.

   • Embed this notice
     Eduardo Mercovich (edumerco@social.coop)'s status on Wednesday, 11-Sep-2024 08:33:35 JST Eduardo Mercovich

     in reply to

     @ntnsndr
     Seems great! What model are you using? :)

   • Embed this notice
     Nathan Schneider (ntnsndr@social.coop)'s status on Wednesday, 11-Sep-2024 11:34:49 JST Nathan Schneider

     in reply to

     • Eduardo Mercovich

     @edumerco Playing with Minstral and Falcon at the moment.

   • Embed this notice
     Eduardo Mercovich (edumerco@social.coop)'s status on Thursday, 12-Sep-2024 00:55:44 JST Eduardo Mercovich

     in reply to

     @ntnsndr
     And what are your thoughts about the experience up to now? Is it useful? How? :)

   • Embed this notice
     Nathan Schneider (ntnsndr@social.coop)'s status on Thursday, 12-Sep-2024 01:29:14 JST Nathan Schneider

     in reply to

     • Eduardo Mercovich
     • Sam Whited

     @sam @edumerco Agreed on the training question—these are standard frontier models as far as I know, so vulnerable to all the problems with energy etc. of the others.

     I find is more or less as useful as cloud-based chatbots, though I feel more willing to share private data with it.

   • Embed this notice
     Sam Whited (sam@social.coop)'s status on Thursday, 12-Sep-2024 01:29:16 JST Sam Whited

     in reply to

     • Eduardo Mercovich

     @edumerco @ntnsndr how are those trained? The biggest problems (in my opinion) with AI aren't really sending stuff to the cloud (which is already a problem be it AI or most non-AI tools that accomplish similar tasks), but the massive environmental impact (mostly created during training, though some part of that is cloud based to) and unethically harvested training data. Those won't necessarily change just because a copy of the AI is running locally instead of on a supercomputer somewhere.