As part of the KDE Plasma 6.5 release, we also got a new release of Plasma Mobile. As there’s a lot of changes, improvements, and new features in Plasma Mobile 6.5, the Plasma Mobile Team published a blog post to highlight them all. The biggest improvement is probably the further integration of Waydroid, a necessary evil to run Android applications until the Plasma Mobile ecosystem manages to become a bit more well-rounded. Waydroid can now be managed straight from the settings application and the quick settings dropdown.
Furthermore, the lockscreen has been improved considerably, there’s been a ton of polish for the home screen and the user interface in general, the quick settings panel can now be customised to make it fit better on different form factors, the first early test version of the new Plasma mobile keyboard is included, and so much more. This is definitely a release I would want to try out, but since I don’t have any of the supported devices, I’m a bit stuck.
This is, of course, one of the two major problems facing proper mobile Linux: the lack of device support. It’s improving due to the tireless work of countless volunteers, but they’re always going to be swimming upstream. The other major problem is, of course, application availability, but at least Waydroid can bridge the gap for the adventurous among us.
re: “problems facing proper mobile Linux: the lack of device support”, I want to make a somewhat offtopic remark, but still within the spirit of this platform: “agentic” coding tools will prove their worth when they will be able to effortlessly write drivers, allowing any hardware to work under any OS.
FriendBesto,
I agree that this is roadblock has proven stubbornly obnoxious for fans of the bring-your-own-os model.
To the extent that the hardware is known to be unlocked (ie doesn’t prevent the owner from flashing their own OS), then support is at least theoretically possible and the problem is one of developer resources to support every specific model. I do think AI could help automate this process. I wouldn’t place too much faith in a generic run of the mill LLM, but I do think a specially trained AI could do much better: reverse engineering OEM firmware and writing a compatible FOSS toolchain as output, ideally compatible with a mainline linux distro. It would be a great use of AI IMHO.
Also, once this AI is available, would the problem be “solved”, or would manufacturers be determined to respond with more device restrictions such that it’s always going to be a cat and mouse game to support devices?
It’s not just phones either. I’ve had to throw out scanners/printers/video capture boards/usb peripherals/smart speakers/smart IOT devices/etc on account unsupported devices & drivers. Open source would be the solution to all of this if you can find FOSS friendly hardware, but there-in lies the rub: so much commodity off the shell hardware has no FOSS support. So training an AI to both reverse engineer the drivers as well as jailbreak the hardware would be tremendously valuable for all sorts of FOSS applications. It would be interesting to throw my hat in the ring, but I wonder how much it would cost to pull off, I’m guessing quite a bit. And there are the usual questions over whether I could make back my time/money, which can be hard to do with FOSS if the work isn’t sponsored.
Alfman,
Agentic AI for coding depends on patterns. They are really good at some tasks, like
“Please start a new React project with TypeScript, add stubs for a user database, backed by OAuth2, and upload all to my personal GitHub account”
But if you have a very niche piece of code, the best they can do is give you advice, and summarize other existing code.
“Help me figure out how Linux handles NE2000 drivers, and why do they have this specific initialization code”.
But they cannot help with:
“I have no sample code, I don’t have the datasheets, and nvidia has a strict DRM on drivers. Can you help me write a performant OpenGL implementation for my card”
It is all about publicly accessible patterns
sukru,
I’m still not sold on programming via english specifications being the best way to use AI for programming. Of course modern LLMs are making all the headlines and have stolen the thunder of AI more broadly, but I still think there’s a whole lot of ground outside of LLMs for AI to cover.
For example there’s a huge difference between commanding an AI like chatgpt to play chess like a pro versus actually using a specialized AI NN trained to play chess proficiently. It’s the same with programming. LLMs are great at interfacing with humans, but they’re not the best at solving problems and I think we’d be gimping the future of AI by reducing everything to LLM solutions. LLM is just one AI tool, and not the best one for every task.
Having an AI that would meaningfully respond to these kinds of english commands would be neat, but IMHO this approach could be an irrelevant distraction to the task at hand (especially if it doesn’t work). I think the key will be to training the AI more directly to do the task at hand, not through spoken commands, but with actual training. The data may not exist in a “public documented form”, but it IS there in a form that’s well suited to computers: binary code. There’s no reason in principal this data can’t be used by AI. Humans can reverse engineer all kinds of stuff, it’s just very tedious. This CAN be automated, but we have to train AI how first. Sure it’s more work than asking an LLM to do it, but it also has a better chance of working. I Personally think this could prove more productive for the types of computing problems we’re trying to solve than LLMs are.
This is something that I’d be interested in working on, but I am guilty of taking on projects that take me away from the work I need to be doing to pay the bills. I’ve long felt that I could accomplish a lot more if I wasn’t held back by the need to earn money, haha.
Alfman,
The current transformer architecture is a massive pattern recognition engine, which can also follow step by step instructions (that is how they can do inductive reasoning).
However they are very far from being able to reverse engineer software, black box, or even with source.
They are still extremely useful. However those “soft skills” and undocumented system hacking by trial and error (which requires usually some hardware tinkering as well), is still out of question.
sukru,
That’s kind of my point though, it’s the wrong type of AI to use. Just because you can ask a transformer architecture to solve a problem doesn’t mean it’s the best approach. Reverse engineering software isn’t some kind of unobtainium, but this goes about it the wrong way.
What you are saying is right about AI models that would rely on having documentation to be able to output code, but having documentation isn’t a prerequisite for training a NN to do a task. And while AGI capable of solve any problem we ask of it in plain english may be the big pie in sky goal everybody else seems to care about most, I actually believe there are a ton of problems including this one where domain specific NNs could make huge strides without us achieving artificial general intelligence first. For better or worse, the AI companies that have all the resources in the world are highly focused on general purpose AI: jack of all trades master of none. But that approach has got a long way to go before it will beat specially trained AI models at technical tasks like reverse engineering.
Unfortunately any work on an alternative mobile OS is pointless as long as certain apps use hardware attestation to ban alternative OSes for “security”
Magnusmaster,
The problem is, security is a valid concern. That makes it difficult to push against, even though it is also a very convenient excuse.
“You are doing this to close off other systems”
“No, we only care about your well being”
sukru,
It’s a valid concern only if the owner is in agreement, and to be fair many owners do. But when security is used as against the owners will it looses legitimacy. “Security” against owners is a fraudulent concept, more apt would be to call it what it the jail that it is.
Of course there’s no issue at all if the measure are voluntary, but as manufacturers start intentionally depriving owner control over the objections of the property owner it becomes a bald face lie.
Alfman,
Unfortunately for us…
We are either not the customers (“the infamous: you are the product”)
Or…
The bank really does not care if 1% of their customers wants a special, less locked down version of their app as an APK.
sukru,
Yes, I agree with you on this. I was just pointing out why their canned “No, we only care about your well being” responses are disingenuous.
“Security” in the abstract is a good thing, but it’s only legit if it serves the owner who’s the only one who should have rights on a device. The unfortunate thing is that “security” has gotten mashed into a bait and switch whereby manufactures omit real security features that empower owners and replacing them with owner jails and calling that “security” instead.
Would be nice for them to release something like KDE Linux, but for mobile phones. For now I am keep wondering how the heck am I suppose to even try Plasma Mobile? I do have Fairphone 6, but there’s no port even for this hugely accessible and developer-friendly phone. Instead, there seem to be PostmarketOS for ancient phones, where you can install Plasma Mobile, OR Mobian that works on nobody knows what, because its website is chaotic, like all Debian-related documentation. Last option on Plasma Mobile website is openSUSE Tumbleweed. I use Tumbleweed on my PC, would like to try on mobile phone, but there is no real documentation for it, also very scarce and chaotic pseudo-documentation next to none. It’s all a mess at this point. I can become your tester, just make it testable and approachable. *sighs*
postmarketOS appears to have made significant progress on a port for the Fairphone 6, which I think is very impressive for such a new device. Short of the device manufacturers themselves shipping their own mobile Linux distro, I think this is better than I would have expected for such a small community. The pmOS folks seem to be doing a lot of great work, but the bottleneck will continue to be hardware manufacturers that don’t provide more support.
The best thing you can do as someone who bought a Fairphone is tell them where you want your money to go. Tell them you want to run Plasma Mobile on your Fairphone 6.