It's interesting to imagine an "average human on the street" being put in the position that an LLM is in. If you asked someone on the street to summarise a page of React Native documentation for you, they'd probably look at you strangely and say "sorry I have a bus to catch, good luck!"
But if you had someone in a room, sat them down with a pen and paper and instructions along the lines of:
- answer the question on the page without diversion
- do not ask for additional clarifying information
- provide your answer in 3-5 paragraphs with a summary
and didn't let them leave until they'd completed the task, then
.. I'm sure the results would be "worse", and much more variable, than GPT output, while being conceptually similar.
Of course, you don't ask a random person on the street to do this, probably because of social constraint, inconvenience, and the expectation that they will not do a better or faster job than you will yourself.
But if you were stuck with a stranger, began to converse, and asked them e.g. "should I get a divorce?" I imagine the conversation would go very differently. If they weren't using their agency to end the conversation as quickly as possible, I imagine there'd be back and forth asking for more detail, exploring the context, sharing their own experience, etc.
I'm rambling, but I just find it very diverting to compare not the output of LLMs versus the output of humans, but the context and situation.