Trust in AI is about the 2%, not the 98%!

I have a dear odd friend. Let's call it Miss Piggy. Miss Piggy vacuum the vacuum cleaner. Yes, Miss Piggy has a extra vacuum cleaner to vacuum the vacuum cleaner Miss Piggy normally use.

I have asked four different AI services(LLM) to create a picture of Miss Piggy vacuum the vacuum cleaner. The result was not as hoped. Why this result? The question is not understood? (the questions is tested in multiple ways and detail level). My bet is not in the training data as (almost) always!

This was a innocent test. Miss Piggy is real. Miss Piggy's odd perk is real. But it showcase the importance of the 2% to be trusted.

Why the importance? A real example with real impact - Cruise (vendor of autonomous cars) didn't detect children as well as it detects adults!