I asked some LLMs to craft some meaningless but beautiful word salads for me. Then tested many LLMs with those word salads (just gave them the word salads as prompts). Every single tested LLM praised the passage and then gave an in-depth analysis of the "beautiful poem" they had received! Reminded me of the critiques of many post-modern or modern paintings that praise pieces of ***.
Update: Till now, only Claude Sonnet 3.5 has been smart enough to catch me red-handed! (23.6.2024).
Then I told the AIs that they were fooled by word salads created by their mates. Their responses were really impressive.
Think we can use this word salad challenge as a new (or if not new, as a good) way of testing LLM intelligence and understanding. Also the ability of LLMs in creating good-enough word salads was impressive.
---------------------
Edit 12 August 2024: Even asking LLMs to create word salads is itself a good way of weighing their relative intelligence. Some are much better than others.