Research map / Large language models / Language, safety and interpretability
Assessing the Capabilities of LLMs in Humor: A Multi-dimensional Analysis of Oogiri Generation and Evaluation
Ritsu Sakabe, Hwichan Kim, Tosho Hirasawa, Mamoru Komachi
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| prompt LLM to write serious instead of humorous response (ablation) | GPT-4.1 · Oogiri human ratings | Overall Funniness (0-4): 1.621 -> 0.976 | Hurt |
| LLM-generated Oogiri responses vs human tiers | Gemini 2.5 Pro · Oogiri human ratings | Overall Funniness (0-4): 1.803 vs Mid 1.913 and Low 1.232 | Mixed |
| LLM responses compared with high-tier human responses on Empathy | LLM generation · Oogiri human ratings | Empathy (0-4): 2.615 (High) -> 2.053 (Gemini 2.5 Pro) | Hurt |
| use LLM as judge of humor instead of humans | GPT-4.1, Gemini 2.5 Pro, Claude Sonnet 4 · Oogiri Overall Funniness | Spearman correlation with humans: 0.169 to 0.266 | Hurt |
| LLM judges rating irrelevant responses | GPT-4.1 judge · Unrelated responses | Overall Funniness (0-4): 0.681 (humans) -> 2.411 (GPT-4.1) | Hurt |
| LLM judging LLM-generated responses (self-preference) | GPT-4.1 judge · Oogiri generation | Overall Funniness (0-4): 1.621 (humans) -> 2.895 (GPT-4.1) | Hurt |
| LLM judges rating serious non-humorous responses | LLM judges · Oogiri serious responses | human-LLM agreement | Improved |
| LLM judges ranking human High/Mid/Low tiers | GPT-4.1, Gemini 2.5 Pro, Claude Sonnet 4 · Oogiri human responses | order accuracy (%): 50.9% to 54.1% | No clear effect |
Closest papers by meaning
- Let's Think Outside the Box: Exploring Leap-of-Thought in Large Language Models with Creative Humor Generation (CVPR 2024 · similarity 0.89)
- Talk Funny! A Large-Scale Humor Response Dataset with Chain-of-Humor Interpretation (AAAI 2024 · similarity 0.88)
- Cracking the Code of Juxtaposition: Can AI Models Understand the Humorous Contradictions (NeurIPS 2024 · similarity 0.87)
- Innovative Thinking, Infinite Humor: Humor Research of Large Language Models through Structured Thought Leaps (ICLR 2025 · similarity 0.87)
- HUMORCHAIN: Theory-Guided Multi-Stage Reasoning for Interpretable Multimodal Humor Generation (CVPR 2026 · similarity 0.87)
- Humor in AI: Massive Scale Crowd-Sourced Preferences and Benchmarks for Cartoon Captioning (NeurIPS 2024 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
