Do you know better?

All AI models fail to solve this simple puzzle

"How many sisters does Alice's brother have?" the researchers wanted to know. The answers are truly curious.(Bild: Anna Velichkovsky)

AI can do everything, knows everything and is therefore superior to us - at least that is the common belief. The fact that this is not (yet) the case is proven by a simple puzzle that causes all the language models on which ChatGPT and the like are based to fail miserably. Would you have known better?

It is a "simple, short, conventional, generally understandable problem" that "can easily be solved by humans" - and yet the common large language models (LLMs) that form the basis for artificial intelligence-based chatbots such as ChatGPT all fail to solve it. And not only that: they even insist on their incorrect solutions, which demonstrates a "dramatic collapse" of their function and logical thinking abilities.