Conversation

Notices

1. Embed this notice
   mhoye (mhoye@mastodon.social)'s status on Sunday, 08-Jun-2025 00:30:36 JST mhoye

   Without tagging in the writer, because I think that would be unfair: there are real perils of language in talking about stochastic repetition machines. This well informed, well intended and written piece of criticism _still_ falls into the reflexive description of ML processing as “thinking” even as it’s making the point that you shouldn’t anthropomorphize the software.

   Software doesn’t think, it only processes.

   • Embed this notice
     Rich Felker (dalias@hachyderm.io)'s status on Sunday, 08-Jun-2025 00:31:15 JST Rich Felker

     in reply to

     @mhoye What the model assigns high likelihood as what you want to hear.

   • Embed this notice
     mhoye (mhoye@mastodon.social)'s status on Sunday, 08-Jun-2025 00:36:41 JST mhoye

     in reply to

     • Rich Felker

     @dalias A response that the model calculates best aligns your future actions with its business model.

   • Embed this notice
     Rich Felker (dalias@hachyderm.io)'s status on Sunday, 08-Jun-2025 00:37:43 JST Rich Felker

     in reply to

     @mhoye Well yes, that's the output of "LLM + alignment". See also: the EKIS protocol.

     For an unadulterated LLM tho it's closer to what I said.