LLMs can predict humor preferences of other models, study finds

A new research paper explores whether one large language model can predict the humor preferences of another using a Cards Against Humanity-style task. The study pitted GPT-4o against Claude Opus 4.5, finding that while simple instructions to mimic the other model yielded minimal improvement, providing behavioral evidence of the other model's choices, especially with rationales, significantly boosted accuracy. This suggests a form of theory-of-mind-like behavior where models can adapt their responses based on observed preferences, rather than internalizing them. AI

IMPACT Demonstrates LLMs can exhibit theory-of-mind-like behavior, potentially improving agent collaboration and personalized AI interactions.

RANK_REASON The cluster contains a research paper detailing an experiment on LLM capabilities. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• alphaXiv
• arXiv
• Cards Against Humanity
• CatalyzeX
• Claude Opus 4.5
• CORE Recommender
• DagsHub
• Gotit.pub
• GPT-4o
• Hugging Face
• ScienceCast

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →