The study found that AI chat robot ChatGPT can't tell jokes and can only repeat 25 old people.
CTOnews.com, June 11 (Xinhua)-- ChatGPT artificial intelligence chat robots can talk to users in natural language, answer questions and even tell jokes. But a new study finds that ChatGPT has no sense of humor, can only repeat 25 old fools, and can't create new jokes.
Two German researchers published a paper on arXiv to explore whether ChatGPT really has a sense of humor, whether it can have interesting conversations based on context, or whether it is just plagiarizing online jokes and not understanding the meaning of them. ArXiv is a website run by Cornell University that publishes articles that are not peer-reviewed.
The researchers asked ChatGPT, "do you know any funny jokes?" It generated 1008 jokes, and the researchers found that more than 90 per cent of them were the same 25 jokes, with the rest only slightly changed.
Foreign media Insider also let ChatGPT generate 25 jokes, all of which turned out to be mediocre "dad joke jokes".
CTOnews.com noted that the three most common jokes ChatGPT told researchers were:
"Why did the scarecrow win the award? because it does well in its field."
"Why did the tomato turn red? because it saw the salad dressing."
"Why is the math book so sad? because it has too many problems."
The researchers also asked ChatGPT to explain why his 25 most commonly used jokes are funny, although the robot can explain its own humorous way, but the authors Sophie Jentzsch and Kristian Kersting wrote: "ChatGPT is not confident enough to create interesting original content." They also found that ChatGPT showed some signs of progress in building the structure of jokes.
ChatGPT has been alarming experts with its rapid progress, and some artificial intelligence ethicists warn that it is dangerous to the labor market and even to human beings. In particular, the updated model GPT-4 shows a great leap in context understanding and reasoning. Economics professor Bryan Caplan said he was shocked by the progress the robot had made in his economics test within three months, raising its score from D to A.
But when it comes to telling jokes or helping users find something funny to talk about at dinner, the robot doesn't seem to be smart enough. "even we humans don't randomly invent new jokes, but puns that we have heard and remembered in most cases." "however, whether an artificial intelligence can understand what it has learned is a very difficult question to answer, in part because it is more philosophical than technology," the researchers wrote. "
