Conversation

Notices

1. peacememories (peacememories@chaos.social)'s status on Friday, 24-May-2024 10:46:44 UTC peacememories

   • ar.al🌻
   • Cheeseness

   @Cheeseness @aral I couldn't find any mention on their website as well, but the summarization feature made me suspicious, so I dove into the code: https://github.com/StractOrg/stract/blob/e74978c5bc1570e10e8f54e2a36a28461f576ead/crates/core/src/summarizer.rs

   Of course, *providing a summarizer* is way less problematic than what other search providers have been doing. But the obvious inaccuracy of these summaries combined with the questionable ethics of anything ChatGPT are imo not compatible with the mission of an open, accurate, and fair search.

   • Wonderskeptic (andersbiork@mastodon.social)'s status on Friday, 24-May-2024 10:46:43 UTC Wonderskeptic

     in reply to

     • ar.al🌻
     • Cheeseness
     • Mikkel Denker

     @peacememories @Cheeseness @aral Maybe @mikkeldenker can clarify the use of AI in Stract?

   • Mikkel Denker (mikkeldenker@mastodon.social)'s status on Friday, 24-May-2024 12:21:16 UTC Mikkel Denker

     in reply to

     • ar.al🌻
     • Wonderskeptic
     • Cheeseness

     @andersbiork @peacememories @Cheeseness @aral Thanks for the tag! The model in production is also Mistral which we host on our own server using vllm. The only reason the code references openai is because everyone seems to have settled on that as the standard api.

     The summary only shows a couple of lines, so I think it can be seen as an expensive snippet that is only computed on-demand when the user requests it. If one wants more information they need to visit the original site.

     [continued]