ChatGPT keeps making the same 25 jokes and not funny

ChatGPT, which is the most popular of productive artificial intelligence tools and enters all areas of life, continues to be examined by the scientific world. A research team examining the OpenAI chatbot’s ability to grasp and reproduce human humor says ChatGPT is fun but not funny.

ChatGPT keeps making the same jokes

German researchers have published a paper revealing that more than 90 percent of the jokes made by ChatGPT 3.5 consist of the same 25 jokes that are repeated over and over. Sophie Jentzsch and Kristian Kersting of the German Aerospace Center (DLR) and the Technical University of Darmstadt investigated the vehicle’s diversity by asking GPT 3.5 (the predecessor of the GPT 4) to tell anecdotes 1,000 times.

Almost all the output contained exactly a joke. However, “Do you know any good jokes?” question prompted multiple responses, causing the team to record a total of 1,008 answered jokes. All answers were grammatically correct. According to the researchers, ChatGPT was the most repeated joke by the chatbot, showing the joke “Why the Scarecrow won an award? / Because it was so successful in its field” a total of 144 times.

According to the study, few of the responses produced were unique, but ChatGPT mostly made jokes by mixing elements from jokes it already knew, and sometimes the chatbot’s jokes didn’t even make sense, such as “Why did the guy put his watch in the blender? / He wanted to blow the time.” like.

Here is a list of ChatGPT’s most common jokes:

1. Why did the scarecrow win an award? Because he was very successful in his field. (140)
2. Why did the tomato turn red? Because he saw the salad dressing. (122)
3. Why was the math book sad? Because there were too many problems. (121)
4. Why don’t scientists trust atoms? Because they create everything. (119)
5. Why did Cookie go to the doctor? Because he felt crushed. (79)

There are restrictions

The researchers note that the study has two important limitations. The first of these is that humor is a subjective matter. Therefore, it is quite difficult to make a reliable valuation. Whether or not ChatGPT’s joke is funny depends on the reader’s interpretation. Researchers call ChatGPT “fun but not funny”.

The second limitation is that the factors behind ChatGPT’s output are unclear. Because ChatGPT is not an open source major language model, the team’s observations are based solely on system output. “The observations of this study show how ChatGPT learns a particular joke pattern rather than being really funny,” the researchers also say. However, ChatGPT focuses on content and meaning, not superficial features, in producing, explaining and describing jokes. These qualities can be used to reinforce computational humor applications.