Comment by denysvitali
10 days ago
I just did something similar to bring mainline Linux to my Daylight DC-1 [1] [2], although possibly more difficult since I had to start from scratch.
I can guarantee the same amount of work, without AI, takes probably 10x or more time (I did it in the past w/ the Google Pixel C [3]).
Unfortunately it looks like pmOS is against AI [4], so I'll likely switch away from pmOS and never contribute a single line in the future. Too bad, I was a huge fan [5].
[1]: https://daylightcomputer.com/product
[2]: https://x.com/DenysVitali/status/2091238391710888416
[3]: https://github.com/pixelc-linux/documentation
[4]: https://docs.postmarketos.org/policies-and-processes/develop...
Sometimes I have to log in just to ask:
Why would you think one of the more beloved projects for linux mobile phones would be pro-AI?
Because one of the biggest hurdles to porting the Linux kernel and device-specific software to more hardware and making existing hardware work better is one of the things that "AI" can (seemingly, caveat) do surprisingly well?
Not to say PM (or any other project) should/shouldn't use AI. There are plenty of other salient reasons to reject AI. But given (from what I remember) postmarket's goal is basically saving as many phones from becoming ewaste as possible, well, an LLM could actually be a formidable way to get more hardware than ever running.
This.
Why would you waste 2 years to port Linux to an old / a device when you can literally do it in ~10 days and save many devices from turning into e-waste?
I really don't get their approach. Besides, Linux (and Linus) accepts and are doing AI contributions, so their argument makes no sense (since they're a Linux distro)
5 replies →
Couldn't these concerns be addressed through two potential angles ?
- You make your AI do all that work, pay xxx tokens worth to get something running and you can use your old phone while taking full responsability for your vibe code. Each user will have to do it, but the whole path could be optimize LLM side as well, after all you're just renting it.
- You use AI to find a working solution but vet it and write everything's needed to meet the submission standard. You put the time needed for it to be accepted as non AI, and proudly put your name on the line as the final author of everything you submit.
Except nothing about AI produces properly open source code? You can't prove it didn't pull from closed source code, no more clean room approaches if everyone just wants to bow to the LLMs.
Even beyond all the other ethical issues I don't see how the open source world accepts this.
2 replies →
In your case, it seems you had an automated (semi-automated, maybe?) feedback loop. In my case, I was the feedback loop :D which ended up being the bottleneck for some parts...
Yes! Removing the human in the loop speeds up the process. Models are better than the average SW engineer nowadays, so this isn't _that_ unsafe... and it really speeds things up