Two years ago, when I was just starting out with Rust, I began working on a proof of concept (POC) for a radio alarm clock.
This week, I decided to pick up the project again and wanted to test Claude AI in parallel.
I had already used this AI to debug some key issues I was encountering with LibreEM5 and Mobian.
Specifically, using the camera with Firefox and the echo with Pipewire.
I was very surprised by what it suggested for the application code, which was functional within 48 hours.
It provided an excellent UI/UX and went beyond what I had asked for (using SQLite and implementing a window to stop an alarm).
All this while retaining the systemd timers I had requested, necessary for launching the application when the phone is in standby mode (which helps with waking up), and using the Radio Browser API.
It’s not perfect yet, but I have a foundation to work from.
First, I’ll address the issue of using my phone with my Hyundai car.
If you’re interested in seeing the application’s output, the code is here:
That is highly important in my opinion. Not just for potential users, but also for AI-lovers, because if new models are trained by your project, projects degenerate and so no further and better “Claude” can be developed.
And how are you able to license it with GPL3 if you don’t know the code licenses that were used to train this AI?
I don’t rate the ethical perspective for personal use case projects (as long as they don’t get merged into real free software projects).
Not just that - in many (most?) jurisdictions, AI output isn’t even copyrightable as it’s something explicitly reserved for human-authored works, so it can’t be licensed at all (though based on that alone you won’t violate any copyrights if you pretend to license it anyway).
I don’t agree or disagree here, because there is still a lot of stuff completely unclear. I mean, you could rewrite an existing program with with restrictive licenses using AI (that is may even trained by the original project) and now we have fully copyright-free code!? That causes a lot of issues and courts just have started to look at this topic.
I am in favor of AI and that AI replacing all human programmers, however we must value human-written code, which in my honest opinion, is very slow, buggy, killer.
In my honest opinion, i do not believe that humans are built for this purpose of coding without pushing themselves to the limit as full stress.
Funny that especially you think this way, because AI has a lot of issues that are fully anti free software. The current AI bubble fight happens also to kill all other AI companies to become the final gate keeper over the whole code that AI can generate in future.
And even more funny, if AI is trained by AI written code (which is a requirement in your dream scenario), AI degenerates. Without new code written by human, AI cannot become better and cannot stay up to date to the newest technology.
You probably should look how AI is functional in detail. There is a lot of misconception in public media and mainstream knowledge.
Of course, “GNU” AI becoming “perfect”. Please, check out some talks of university profs that actually research on the abilities of this technology. What you dream about is understandable, but far far far away from reality. Not a matter of time and/or energy, a matter of possibility of this kind of technology.
To have some form of “AI” that can fully replace all developers, it requires completely fundamental different approaches. And how do you want to verify 100.000 new lines of code that are written in one day is it still free software if there is so much new code that it is impossible to review it, because next day another 100.000 lines are changed?
Please, check out some talks of university profs that actually research on the abilities of this technology. What you dream about is understandable, but far far far away from reality. Not a matter of time and/or energy, a matter of possibility of this kind of technology.