Daily Linux & Open Source News - S01E73 - Fedora goes all in on AI
Added 2024-04-11 08:05:08 +0000 UTCHey everyone!
In this one, we have:
00:06 Small browsers seeing growth in the EU
03:21 Proton buys Standard Notes
05:14 Fedora wants to go all in on AI
https://communityblog.fedoraproject.org/strategy-2028-april-2024-update/
Have a nice day!
Comments
I don't understand why we keep blurring the difference between an operating system and the applications it runs. At best, AI should be applications with many competing providers. Users can decide what tools they want to use or not. They shouldn't be baked in.
Steve C
2024-04-11 16:02:27 +0000 UTCAI feels like a FOMO virus that trades foundations of good human life for trinkets and I'd rather it disappear from the Earth. BUT, if there's any group of people I'd trust to handle it as well as possible, it's the open source community. If there's any group whom I'd trust to pull off the execution competently, it's Fedora. Plus they're usually on the forefront of technologies anyways, it's in-character for them to be first to make / use the new thing. I might disagree with the application of AI system-wide vs app-by-app though? App-by-app is the most conservative way to do it, it's just used for some isolated features in apps that feel like it, but I feel like that as the default could be even worse for infiltration into the ecosystem, inconsistent experiences, randomly managed data and security practices, and increased difficulty in running locally because each app's individual AI models could take 0.5 to 16 GB, and that'd destroy disk space even if they could share a system-wide model. I think having one local model of each type (text-to-text, text-to-image, image-to-text, text-to-audio etc.) that can be used at a system level via a GNOME extension or KDE widget and/or separated app, and requested for via a GTK or Qt or Flatpak permission-restricted API, would be best. And individual apps are free to ship their own models instead, but I think managing them at a system level would give the most coherence and manageability, and dare-I-say it usefulness if your per-app AI experience is based off of your entire computing experience (e.g. Talk to it about what you wrote yesterday, what you want to draw etc., having a personal history context). It's maximally spyware-y if such an integrated experience is omnipresent and can/does report back to a company at any time, but a least-privileged, local-first, transparent system from the Linux community is one I could trust. Just some thoughts 🤷.
NotMyName
2024-04-11 13:06:36 +0000 UTC
