> AI itself could probably discover better architectures than transformers (willing to bet heavily on this)

Is there any existing cases of LLMs coming up with novel, useful and namely better architectures? Either related to AI/ML itself or any other field.