There are almost no real world tasks that LLMs outperform humans on, operating by themselves. Pair them with a human for adaptability, judgement, and real world context and let the human drive, sure. Just let it loose on its own? You get an ocean of slop that doesn't do even close to what it's supposed to.