2. I would indeed see a lot of benefits with only allowing self-hosted genAI. I wonder how practical it is to enforce that, and such a policy would IIRC be first of its kind, but I’d like to see it; I won’t dispute this.
3. those are about use of AI to replace critical thinking ( resulting in stagnating critical thinking, not coding. to show the point you want to make, you’d need to find a study about the effect of LLM use on code quality—not learning, code quality, since we’re talking about seasoned KDE devs here.^[1]
4. I don’t like this debate because “just a tool” is not a very defined phrase. How I see it, it means AI needs to be used carefully and using it lazily is misuse that results in dangerous things, but the term’s also too often carelessly misused to mean very dangerous things (ironic?), so I see the reason in arguing against it and the meaningless derived from lack of agreed meaning. Let’s just say it’s powerful (💥) and call it a day.
5. I would’ve said this was a good point a year or two ago. the legal environment is clearly not heading thataway (plus, politically, I don’t think property should exist.)
6. Combing through, identifying, and only copying public domain portions of the code is too much of a hassle that they’d use an LLM to do it. Public domain for everybody!

[1] I found https://link.springer.com/chapter/10.1007/978-3-031-83911-5_8 and https://link.springer.com/chapter/10.1007/978-3-031-90573-5_7. The first some seems awesome and finds that ChatGPT code for small, one-function–type tasks (somewhat more pragmatic than Leetcode but only somewhat) was better than human code from GitHub in aggregate. I would wait for more studies before going to conclusions. The second one says “delves into” in the abstract and has 32 pages so even though I’m guessing it wasn’t AI-generated I ain’t reading allatritenow