Conversation

Notices

1. Embed this notice
   tech? no! man, see... (technomancy@icosahedron.website)'s status on Friday, 05-May-2023 08:54:27 JST tech? no! man, see...

   in reply to

   • Andrew Plotkin

   @zarfeblong I think it's a bit of a stretch to claim that this is the missing piece that allows LLMs to be used ethically

   the leaked post is mostly about advances in Stanford's Alpaca, which is a series of refinements which are layered on top of Facebook's Llama model

   they're saying that it's possible to take an existing huge model (which was illegally leaked) and make incremental modifications to it to the point where it will be competitive with state of the art megacorp models, but since it's based on Facebook's model, it's still impossible to answer questions of attribution

   so the specific systems listed in that article don't really get us any closer to ethical systems

   • Embed this notice
     Andrew Plotkin (zarfeblong@mastodon.gamedev.place)'s status on Friday, 05-May-2023 08:54:32 JST Andrew Plotkin

     in reply to

     This isn't entirely negative. It opens up new ethical uses for AI tech! You can train a text or image model *entirely on your own work*, at home -- no need to rely on a predatory tech company *or* questionably sourced data.

     Indeed, I know at least one artist who is doing this right now.

   • Embed this notice
     Andrew Plotkin (zarfeblong@mastodon.gamedev.place)'s status on Friday, 05-May-2023 08:54:34 JST Andrew Plotkin

     This went around today:

     https://www.semianalysis.com/p/google-we-have-no-moat-and-neither

     Leaked Google document, basically saying "We have no competitive advantage over open-source LLM work."

     This is interesting Google-kremlinology. But surely the important point is that AI -- at least the text-generation, image-generation parts -- are out of the hands of tech companies. Anybody can run that stuff at home. It is now impossible to legally regulate or limit it.

     (DeCSS flashbacks, anyone?)