[rekabis]

Eeehhhhhh… while I am cautiously optimistic of AI, everything I have seen to date still makes me want to avoid its creation of code. It still hallucinates far too wildly, over engineers far too eagerly, and can only crib from prior examples; it cannot look at a situation and use judgement and nuance to create truly appropriate and effective code.

So using AI as a guide while you write code, sure. It’s probably the best rubber ducky ever built. But letting it actually write the code? Nah, man. That’s the path to unmaintainable slop crammed full of technical debt.

[Krei-se]

Great differentiation. Having an AI write code means you are not clear about your intentions because if you would be you'd just write those intentions in clear, concise language: code.

Torvalds does not code much himself anymore so i take his stance not so seriously. If you try to push AI generated code the question is not whether it works or is good or whatever, it's that i have no one to ask when problems come up and maintain that code.

Who takes further responsibility? If you merge an AI generated pull request its up to the main maintainers to go ask another machine if they have questions?

[rekabis]

For me, I take a page out of IBM’s book:

> A computer can never be held accountable, therefore a computer must never make a management decision.

Except this should really be “any kind of choice or decision” in general, not just management ones.

This means that AI should never be allowed to make any kind of decision or be able to employ any kind of choice, including what code to create and how the program gets constructed, because it can never be held accountable.