🥶 Linux (and therefore me) is gonne be so fucked in five years when they (like every project on the ai hype train) can’t understand the spaghetti code (including future ai).
Tbh I trusted him up until the point where he compared AI to compilers. That’s just absolutely crutial misunderstanding where the issues with AI usage lies. Comparing using compiler to code, vs using AI to code, is so immensly stupid that I have started doubting him.
Not because I don’t trust in his skill, but because AI is addicting and dealing with additction/risk of psychosis responsibly takes an entirely different skillset. And resorting to toxicity when defending AI with absolutely bonkers arguments, like reducing it’s risk to “ShOuLd We AlSo bAn CoMpIlErS? WrItE iN aSeMbLy!!” is not a good sign that he is handling it well.
And he even compared it to using a fucking text editor to write commit messages, that wanting to limit AI use in commited code is the same as wanting to ban text editor usage. That’s just plain insane misunderstanding of what the issues are.
What!! You would use an assembler to make binaries to run on a general purpose machine code interpreter instead of using an atomic knife and a salt-accelerator to write the instructions on a silicon die?
Albsoluridonculoustrous!
But honestly, if he is saying that, I now have a good reason to assume he can’t be trusted with having proper logic in the kernel.
IIRC one big problem with using AI code in Linux was license of created code. If that is somehow fixed, which is not the case, then it might at least be fine for those coming in the future.
🥶 Linux (and therefore me) is gonne be so fucked in five years when they (like every project on the ai hype train) can’t understand the spaghetti code (including future ai).
Maybe I’m just being optimistic, but I trust Torvalds. The same can’t be said about the typical vibe-coder, though.
Tbh I trusted him up until the point where he compared AI to compilers. That’s just absolutely crutial misunderstanding where the issues with AI usage lies. Comparing using compiler to code, vs using AI to code, is so immensly stupid that I have started doubting him.
Not because I don’t trust in his skill, but because AI is addicting and dealing with additction/risk of psychosis responsibly takes an entirely different skillset. And resorting to toxicity when defending AI with absolutely bonkers arguments, like reducing it’s risk to “ShOuLd We AlSo bAn CoMpIlErS? WrItE iN aSeMbLy!!” is not a good sign that he is handling it well.
And he even compared it to using a fucking text editor to write commit messages, that wanting to limit AI use in commited code is the same as wanting to ban text editor usage. That’s just plain insane misunderstanding of what the issues are.
What!! You would use an assembler to make binaries to run on a general purpose machine code interpreter instead of using an atomic knife and a salt-accelerator to write the instructions on a silicon die?
Albsoluridonculoustrous!
But honestly, if he is saying that, I now have a good reason to assume he can’t be trusted with having proper logic in the kernel.
IIRC one big problem with using AI code in Linux was license of created code. If that is somehow fixed, which is not the case, then it might at least be fine for those coming in the future.
You have good reason to trust him. Trusting the future however is a losing game.
Yeah, there’s a distinction between using an LLM as a tool and letting the LLM be in control.
No, the kernel will be understood or replaced extremely quickly. This is the beauty of open code.