Multi-Agent Comedy Club: Investigating Community Discussion Effects on LLM Humor Generation

2026-02-16

AI Summary

This work addresses the limitation of current large language models (LLMs) in humor generation, which typically neglect dynamic community feedback and rely on static prompts for evaluation. We propose the first framework that integrates community-level discourse into LLM-based comedy creation by constructing a multi-agent sandbox environment to build a “social memory.” This memory captures, filters, and stores user comments and audience reactions, which are subsequently retrieved to condition stand-up comedy generation. Evaluated through 250 A/B tests and expert assessments across 15 dimensions, our approach significantly outperforms baselines with a 75.6% win rate, demonstrating notable improvements in key aspects such as Craft/Clarity (+0.440) and Social Response (+0.422), thereby transcending the conventional static prompting paradigm.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: Generation
• Cognitive Modeling & Cognitive Systems: Computational Creativity

Application Category

• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems
• Social Networks and Social Media: Generative AI / large language models and their impact on social systems

Abstract

Prior work has explored multi-turn interaction and feedback for LLM writing, but evaluations still largely center on prompts and localized feedback, leaving persistent public reception in online communities underexamined. We test whether broadcast community discussion improves stand-up comedy writing in a controlled multi-agent sandbox: in the discussion condition, critic and audience threads are recorded, filtered, stored as social memory, and later retrieved to condition subsequent generations, whereas the baseline omits discussion. Across 50 rounds (250 paired monologues) judged by five expert annotators using A/B preference and a 15-item rubric, discussion wins 75.6% of instances and improves Craft/Clarity ({\Delta} = 0.440) and Social Response ({\Delta} = 0.422), with occasional increases in aggressive humor.

Problem

• LLM humor generation
• community discussion
• public reception
• multi-agent interaction
• stand-up comedy

Innovation

• multi-agent simulation
• community discussion
• social memory
• LLM humor generation
• controlled sandbox