I agree with your opinion. The definition of novelty largely depends on subjectivity. Results produced by combinations of LLMs can also be novel. Combining multiple insights, imitating them, and then organizing them into one can also be considered sufficiently novel. LLMs have a kind of bias due to RLHF, and that may make them uninteresting to humans, but when a "human" uses them, they can enable the human to do more, and that greater amount of work can create new novelty.