That depends… are you talking about actual AI, which is hard, or LLMs, which are resource wasteful, inconsistent prediction engines that even at their best are catastrophically wrong 1 out of every 100 times you use them, and require a $1000 GPU just to get a result in reasonable time?
Even your example there? Can be done in a 50 line bash script, with the exception of the “Review” folder, which would be wonky anyway because the files that you can decide are not useful, you already know how to decide are not useful, and the ones you don’t, you would have as much trouble with as the AI would, so you haven’t really gained anything. Instead of a continually iterating and improving script that does exactly what you want, you get an unpredictable application that will produce different output every single time it runs, with potential for regressions every single time you do it.
Basically, we’re still at the point where anything a general purpose AI could do on a local operating system, you could do better and more efficiently with a purpose-built tool.
But we live in an era where people legitimately put out Electron apps and call them “lightweight and efficient” with no dissonance whatosever, so I’m sure there are a million people ready to argue.
That depends… are you talking about actual AI, which is hard, or LLMs, which are resource wasteful, inconsistent prediction engines that even at their best are catastrophically wrong 1 out of every 100 times you use them, and require a $1000 GPU just to get a result in reasonable time?
Even your example there? Can be done in a 50 line bash script, with the exception of the “Review” folder, which would be wonky anyway because the files that you can decide are not useful, you already know how to decide are not useful, and the ones you don’t, you would have as much trouble with as the AI would, so you haven’t really gained anything. Instead of a continually iterating and improving script that does exactly what you want, you get an unpredictable application that will produce different output every single time it runs, with potential for regressions every single time you do it.
Basically, we’re still at the point where anything a general purpose AI could do on a local operating system, you could do better and more efficiently with a purpose-built tool.
But we live in an era where people legitimately put out Electron apps and call them “lightweight and efficient” with no dissonance whatosever, so I’m sure there are a million people ready to argue.