Recent searches
Search options
The number of people that proposed to "integrate ChatGPT into KDE" with absolutely no detail on how that would look like and what problem it would solve is too damn high
It's also funny (for some definition) of funny how these people imply that AI == ChatGPT
And even assuming there was a solid use case for ChatGPT in KDE there's tons of open questions/concerns about whether this is actually possible in accordance with our values
@nicofee Agreed, as long as ChatGPT is as closed as it is it's going to be problematic.
I do think there could be some interesting use cases plugging some documented KDE API and allowing the user to e.g. ask through krunner "change to dark colorscheme" or "switch wallpaper to a slideshow of my pictures folder", but it's not easy to implement and rely on, and it could very well just be a third party, given how it would be indeed problematic.
It is clearly impossible when it comes to ChatGPT but there are many simple models that can run locally with a small impact on resources like this one, that could be useful for semantic search:
https://huggingface.co/sentence-transformers/all-MiniLM-L6-v2
Other ideas are tasks like “find all the PDFs about Foo and copy them in a folder on the desktop”. Things like this should be possible without big models like GPT4.
Something that KDE software could provide independently from the models are some confirmation dialogs that summarize a task.
In the example above it would be a dialog that says “Move N PDFs into ?" and it lists files with a conventient UI.
These dialogs could be used by any program, not specifically by a ML model, but they makes sense with ML to be sure that the model is interpreting correctly the prompt.
Something that KDE software could provide independently from the models are some confirmation dialogs that summarize a task.
In the example above it would be a dialog that says “Move N PDFs into ?" and it lists files with a conventient UI.
These dialogs could be used by any program, not specifically by a ML model, but they makes sense with ML to be sure that the model is interpreting correctly the prompt.
@nicofee Correct me if I am wrong. KDE is a desktop environment.
Linux is largely independent from the desktop where differences are, for instance, in package management.
Desktop is window dressing, which is nice. But, for the most part we keep going back to text.
From what I see, the most recognized implementation of AI is in programming assistance through and IDE. That would be ChatGPT as copilot.
Why is AI relevant to KDE in the first place?
@johnbessa KDE is not a desktop environment. It's a community of people producing software, including a desktop environment, but also lots of other programs like an email client or an IDE.
There certainly are ways in which AI integration could make sense for those
@nicofee Obviously you just have to ask ChatGPT what problem in KDE it could solve ;)
@nicofee This should be illegal and punishable by death.
@nicofee a lot of the suggestions is automation that's already possible in some form, but somehow needs a LLM to function 