Consciousness recognises consciousness. Every production of a conscious being resonates to some degree with other conscious beings. Consciousness continually reforms and remixes into another entity that has a unique take on reality.
LLM-generated content is so easy to spot precisely because the uniqueness is limited to the prompting chops of the user – but I’m willing to bet that the number of ways in which the majority of humans prompt LLMs is very limited, producing the same exact underlying format – which only changes at the point of training.
Unlike a person, who changes every day, these models are static and therefore the pattern matching kicks in after a little exposure.
Asking me a question a year ago and then again today would yield a different answer in most cases. Go back 10 years and the responses are likely extremely different.
We love great writing and art because it captures fibers of truth from our own experiences of the world – it resonates. LLMs can imitate this of course, but their training inputs are so limited as to hamper their abilities to generate truly novel insights (albeit for the right question they can still provide extraordinarily useful outputs.)