They don't just give different implementations of the same categorical output, but categorically different output. LLMs do not write like humans.