Post

i am more of the view that it's impossible for LLMs to be used for automation myself

automation is: a machine does a series of steps when you press a button. if you press the button and the steps don't get done correctly that means the machine is broken and you fix it if the outcome of pressing the button is always "something different will happen than last time", not automation

with LLMs you push a button and you get a random outcome that you hope is sorta what you wanted. it might be what you wanted, but that will have been by chance, not because a procedure was followed. even if it gets so "good" that it "always" works (never happen) it will not have been automation

Just because a machine can break that doesn’t mean the machine isn’t working up until it breaks AI bad, sure, but this is a pretty bizarre way to try to define automation as a way to “other” AI. Seems a lot more straightforward to just say that AI is bad because it hurts people and the planet