I agree, and I think it’s closely related to something else I dislike about AI — art or other media. The best it can do is interpolate among other, generic, mediocre training data. There are a few cases (novel go strategies, optical illusions) where a human has carefully guided it to a new creative output. But on its own, it’s missing that obsessive need to render some internal idea into the world.
I run into this in programming. I can add the AI agent to do some administrative tasks, like factoring out a React component. But it’s never yet been able to solve a problem I got stuck on, where a teammate quickly identified the extra aspect I needed to take into account, or the way I needed to shift my approach.
AI is great at the instinctual, pattern-matching part. I wish we would use it to eliminate the redundancy in our writing and art, rather than amplify it.
Yeah, this discrepancy really irks me in programming, too. It’s really good at known problems, like student homework or whatever task a middle manager will throw at it to see how well it works.
But because of the nature of software – if there is a solution, you can easily share it with everyone in the world – it’s kind of our job to work on anything but known problems.
Yeah, there’s gonna be some known parts, where it may be able to assist, similar to a library or StackOverflow. But if it can put together your whole solution without tons of human input, chances are that solution is already out there and you should be using it instead.
I agree, and I think it’s closely related to something else I dislike about AI — art or other media. The best it can do is interpolate among other, generic, mediocre training data. There are a few cases (novel go strategies, optical illusions) where a human has carefully guided it to a new creative output. But on its own, it’s missing that obsessive need to render some internal idea into the world.
I run into this in programming. I can add the AI agent to do some administrative tasks, like factoring out a React component. But it’s never yet been able to solve a problem I got stuck on, where a teammate quickly identified the extra aspect I needed to take into account, or the way I needed to shift my approach.
AI is great at the instinctual, pattern-matching part. I wish we would use it to eliminate the redundancy in our writing and art, rather than amplify it.
Yeah, this discrepancy really irks me in programming, too. It’s really good at known problems, like student homework or whatever task a middle manager will throw at it to see how well it works.
But because of the nature of software – if there is a solution, you can easily share it with everyone in the world – it’s kind of our job to work on anything but known problems.
Yeah, there’s gonna be some known parts, where it may be able to assist, similar to a library or StackOverflow. But if it can put together your whole solution without tons of human input, chances are that solution is already out there and you should be using it instead.
It’d be interesting to have an AI look for recurring boilerplate from StackOverflow and suggest new libraries or language features.