In this article, the author explores the concept of humor and the predictability of Large Language Models (LLMs) like ChatGPT. They compare the use of Markov chains and LLMs in generating sentences and jokes, highlighting the importance of surprise in humor. The author argues that while LLMs are great at predicting sentences based on context, they lack the creativity needed for generating humor. They also discuss the limitations of current LLMs in artistic expression and the potential for developing a new type of language model. Overall, the article delves into the intricacies of humor, language models, and the challenges of balancing predictability with creativity.
https://emnudge.dev/blog/markov-chains-are-funny/