A new report indicates that large language models (LLMs) exhibit some agreement in their humor preferences, with an average overlap of about 75%. While all tested models were less consistent than humans in their judgments, Gemini proved to be the most consistent. The study found a general preference among LLMs for absurd humor and roast joke formats. AI
IMPACT Understanding LLM humor preferences could inform more nuanced AI interactions and content generation.
RANK_REASON Research report analyzing LLM preferences. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →