I'm no expert, but I thought that while many modern applications that attempt to claim some of the power of calling themselves "intelligent" and "new"/sleek/modern are mainly frauds, on the other hand the PROPER usage of a generalized learning model can be quite insightful. Plus anyway it's going to get better, so the poor performance issue won't always be true:-).
But that's why I like this (OP) image so much: it points to how lack of training data leads to bad outcomes, more so than the technology itself. Even some humans cannot often recognize such differences, especially those who simply don't care.
And this has VERY bad implications for us all, as A"I" gets deployed in e.g. police departments across the world.:-(:-(:-(
Combined with uninformed people thinking AI implies general intelligence and then criticizing LLM's for not matching these false expectations. It's kind of like criticizing cruise control for not driving you home.