How funny are the frontier AI models? (invertedpassion.com)

Recent experiments at the Lossfunk lab reveal that advancements in AI models are leading to an increase in humor generation capabilities. The study involved six different closed models from OpenAI and Anthropic, such as GPT-6 Astra and Fable 5.1, prompting them to produce original jokes using randomly paired words. Interestingly, these models demonstrated significant improvement in generating humorous content, with GPT-6 achieving over 50% positive reactions (chuckle or laughter) from human raters, indicating a clear trend of models becoming funnier over time. This development is noteworthy for the AI/ML community, as it challenges the notion that humor—a complex and subjective domain—might be distinctly human. The correlation between the ability to produce jokes and performance in other areas underscores that the growth in capabilities within large language models isn’t limited to verifiable tasks like math or coding but extends into nuanced realms like comedy. Ultimately, while AI models are not poised to replace comedians, this progress points to an exciting frontier in machine understanding of creativity and expression, suggesting potential for even greater humor generation with future enhancements.