> This is another attitude I’ve noticed in AI enthusiasts: the AI never fails, the AI is never the problem, it’s always the human who failed the AI by not asking it properly.
I assume you have never used any tool ever? It is common knowledge by now that AI is not a one shot wonder in a lot of cases (weirdly so every argument seems use that as a precursor).
You are saying the equivalent of:
Failed to compile source code - oh shit the compiler is wrong not my source code
If you expect a "stochastic parrot" to be a one shot wonder, surprise you get what you looked for.
Here's a simple thought exercise:
You have a macro for generating a for loop that in X% of cases is correct, what is the X that is needed until it is practically useful.
The answer is not actually that straightforward, because typing it is far more cognitively complex that actually reading and verifying that it is a for loop.
Or another one that is far more relatable, what is the probability of accurate autocomplete that is needed for be useful? This is much easier to answer, as long as it is >0% it is useful for the small hits it provides, because checking is easy.
What does this mean? As long you have a better autocomplete using AI compared to the static analysis autocomplete you are already "better off". Again the simplest example, and at current stage this has long passed.
If you can't shape a tool to your workflow sorry but that's on you. You don't have to use the tool and that's perfectly fine as well. But that doesn't mean people have to hire you if there is someone who can use it and "performs better" (whatever this means).
I assume you have never used any tool ever? It is common knowledge by now that AI is not a one shot wonder in a lot of cases (weirdly so every argument seems use that as a precursor).
You are saying the equivalent of: Failed to compile source code - oh shit the compiler is wrong not my source code
If you expect a "stochastic parrot" to be a one shot wonder, surprise you get what you looked for.
Here's a simple thought exercise: You have a macro for generating a for loop that in X% of cases is correct, what is the X that is needed until it is practically useful.
The answer is not actually that straightforward, because typing it is far more cognitively complex that actually reading and verifying that it is a for loop.
Or another one that is far more relatable, what is the probability of accurate autocomplete that is needed for be useful? This is much easier to answer, as long as it is >0% it is useful for the small hits it provides, because checking is easy.
What does this mean? As long you have a better autocomplete using AI compared to the static analysis autocomplete you are already "better off". Again the simplest example, and at current stage this has long passed.
If you can't shape a tool to your workflow sorry but that's on you. You don't have to use the tool and that's perfectly fine as well. But that doesn't mean people have to hire you if there is someone who can use it and "performs better" (whatever this means).