We know ChatGPT’s database is limited in terms of timeline. But what about humour? How does the language learning model do on the funny scale? A study by a group of students from the Technical University Darmstadt, Germany, found that while the model may be able to understand and replicate learned jokes, it’s still a long way from creating “intentionally funny original content.” What was the experiment? Students Sophie Jentzsch and Kristian Kersting tested ChatGPT’s ‘sense of humour’ by asking the model to generate, explain, and detect jokes. In their paper, the students said they wanted to understand ChatGPT’s capability to grasp and reproduce human humour. What was the conclusion? The study concluded - “ChatGPT is able to capture contextual information and adjust its responses accordingly to the preceding course of conversation. This is an intriguing capacity, which we would like to include in future investigations… [Its] focus bears on content and meaning and not so much on superficial characteristics. These qualities can be exploited to boost computational humour applications.” Why it matters: ChatGPT’s success in November 2022 brought Artificial Intelligence into focus across the globe. However, the model’s system still fails to meet the requirements of “open science” by keeping data, training details and model characteristics private. In this respect, the abovementioned study provides an interesting insight into ChatGPT’s understanding of an important human characteristic – humour. ChatGPT tested on Joke generation, explanation and detection Based on prompt-based experiments, students asked ChatGPT to provide a joke a thousand times, explain each…
