Do you… think it matters? Tool generates code, you look at it, check it and correct it. You send it to a maintainer by giving your name and email, you are now expected to help maintain this code if it gets merged, it has your name on it.
Explain to me why writing the code by hand or with a “deterministic” alternative why would change the quality of what you do or are expected to do in future.
He compared an LLM to a compiler suggesting that the concerns about using them are the same as suggesting that everyone should write in assembly.
The comparison is invalid, since one mode is deterministic, while the other is not.
LLM-driven development leads to programming via the abstraction of the model, disconnecting you from the source to the point where you don’t understand what it’s done or why. He’s suggesting that this isn’t a problem because an LLM is just like a compiler, which suggests that he’s arguing in bad faith or doesn’t understand the “tools” he’s using.
writing the code is the easy part, having a good mental model of what’s going on under the hood and planning ahead is the real challenge. if a chatbot puked out the code it’s easy to assume that it’s right because it worked under your conditions but you might miss stuff that you’d otherwise acknowledge if you wrote the code.
using the chatbots directly discourages thinking and you’re deskilling yourself in the long run, depriving yourself of hands-on experience that you’d usually pick up through your coding career.
having a chatbot generate code for you also usually goes hand-in-hand with writing more code faster, which tends to be bad if you like stable software and want to prevent maintainer fatigue.
and that’s before we get into the ethical and legal problems.
You don’t need to check that the machine code your compiler made matched the code you wrote. You don’t need to check that the commit message an editor sent matched what you typed.
You absolutely do need to check and correct everything that comes out of AI because it’s at best like 80% correct, and that’s if the subject matter is easy.
Do you… think it matters? Tool generates code, you look at it, check it and correct it. You send it to a maintainer by giving your name and email, you are now expected to help maintain this code if it gets merged, it has your name on it.
Explain to me why writing the code by hand or with a “deterministic” alternative why would change the quality of what you do or are expected to do in future.
He compared an LLM to a compiler suggesting that the concerns about using them are the same as suggesting that everyone should write in assembly.
The comparison is invalid, since one mode is deterministic, while the other is not.
LLM-driven development leads to programming via the abstraction of the model, disconnecting you from the source to the point where you don’t understand what it’s done or why. He’s suggesting that this isn’t a problem because an LLM is just like a compiler, which suggests that he’s arguing in bad faith or doesn’t understand the “tools” he’s using.
writing the code is the easy part, having a good mental model of what’s going on under the hood and planning ahead is the real challenge. if a chatbot puked out the code it’s easy to assume that it’s right because it worked under your conditions but you might miss stuff that you’d otherwise acknowledge if you wrote the code.
using the chatbots directly discourages thinking and you’re deskilling yourself in the long run, depriving yourself of hands-on experience that you’d usually pick up through your coding career.
having a chatbot generate code for you also usually goes hand-in-hand with writing more code faster, which tends to be bad if you like stable software and want to prevent maintainer fatigue.
and that’s before we get into the ethical and legal problems.
Slop coders see Kernighan’s Law and think “I can do so much worse”
You don’t need to check that the machine code your compiler made matched the code you wrote. You don’t need to check that the commit message an editor sent matched what you typed.
You absolutely do need to check and correct everything that comes out of AI because it’s at best like 80% correct, and that’s if the subject matter is easy.