@ed @ariadne @kfogel The original training data is absolutely necessary to have an understanding of what the model was created to do on par with its creators. Arguing about how someone might prefer to modify it is disingenuous. Source is also there to study how the software was made, what properties it has (especially trust/security properties), etc.!
Conversation
Notices
-
Embed this notice
Rich Felker (dalias@hachyderm.io)'s status on Friday, 21-Jul-2023 13:17:31 JST Rich Felker
- Haelwenn /элвэн/ :triskell: likes this.
-
Embed this notice
Stefano Maffulli (ed@social.opensource.org)'s status on Friday, 21-Jul-2023 13:17:32 JST Stefano Maffulli
@ariadne I'm starting to think you're biased 😉 go ask around to any researcher in AI, they'll say that the original training data is not strictly necessary.
@kfogel