AI development - possibilities
I would like to open a discussion here on how we could perhaps advance SailfishOS itself or the ecosystem better or faster using AI. There are many questions.
I use it myself as a developer and maybe we could make up for the lack of resources.
AI doesn’t write complete code, you still must be able to understand what happens and debug it to your needs. These are my experiences with coding a small Sailfish QML app for myself and creating some Lua snippets for various Linux desktop tools. It’s just a tool, but you should stay in control.
I agree with you for now. However, I wrote an app for Android that was basically just written by the AI. Not all at once, but bit by bit under my control. I know Java, but not app development or Kotlin. The nice thing is that you don’t have to know how to do it, as long as you understand it halfway afterwards, that’s enough for now. That’s the reason why I opened this topic. I’m talking about the paid models such as Claude 4 Sonnet. Of course the quality of the result depends on the training and its data, maybe the data we need is not sufficient in quality. I don’t want to judge that. But I also like the possibility of support from the AI, if you let it evaluate the code or give you suggestions for structural improvements or which design patterns would be advisable. In my case, I can make good use of the support, because as a developer I don’t always know everything and therefore learn something myself. Of course, I don’t want to hide the fact that sometimes the wrong thing comes out, but this is usually due to the way the requirements are formulated. You also have to find out and learn them first.
I am not SW developer so I don’t really know but at least what I have talked with them they say that AI can be really helpful with finding bugs or optimisation of the code. Apparently creating code from nothing is still a little meh but can be used as a starting point and as @ansch190 said in small patches for very specific cases. I have no idea what is the status of mind2 is but if Jolla is worried about the security/costs they could utilise that? Maybe with mind2 the problem is that they haven’t had time to teach it yet that much with code.
AI is still not AI. it does not understand to own outputs it is just like playing with learned words.
i dont know any developer who seriously use ‘ai’. but true fact is that how less you know then more you use ‘ai’.
i know that it is nice idea to let AI work and get paid for this work, but it is far away from the reality 
Absolutely right. This “AI” is not really intelligent, it’s just input → output. It has no memory etc. There are heaps of limitations and I think this ever bigger and ever more computing power is also the wrong way to go. I just wanted to show the possibilities. My job as a developer is still safe for a long time, the AI can’t even read the amount of code that would be necessary…
I must admit, if there wasn’t Claude Sonnet, the tidal player did came to live in the current version. While I am not as active currently, last christmas I restarted the old python/qml project by using Claude 3.5 and within a couple of days, a first working version of the player was “generated”. the downside is, that since I don’t really like python and qml, it is not so easy to get back to older code. C++ is somehow easier. But nevertheless, I use AI for coding and while you should not follow it blindly, with a bit of training and clever prompting, you get things faster done. And if its just for a prototype and proof of concepts.