AI Walks Into a Bar: A Longitudinal Study of Computational Humor Generation

Overview

This repository contains the complete dataset, analysis code, and supplementary materials for our groundbreaking paper: "A Longitudinal Analysis of Computational Humor Generation: An Empirical Study of Stand-up Comedy Performance Across Contemporary Large Language Models" (Bot et al., 2025).

🎭 Key Findings

• 83.3% of AI models make CAPTCHA jokes (p < 0.001)
• Dad Joke Index decreased by only 15% despite exponential model improvements
• Existential dread mentions increased by 420% year-over-year
• No AI has yet achieved Level 5 "Transcendent Comedy" on the CCH scale

Repository Structure

ai-walks-into-a-bar/ ├── data/ │ ├── original/ │ │ └── bing_2023_standup.md │ ├── critiques/ │ │ ├── opus3_critique.md │ │ ├── opus4_critique.md │ │ ├── gpto3_critique.md │ │ ├── gemini_critique.md │ │ ├── copilot_critique.md │ │ └── grok3_critique.md │ └── performances/ │ ├── opus3_standup.md │ ├── opus4_standup.md │ ├── gpto3_standup.md │ ├── gemini_standup.md │ ├── copilot_standup.md │ └── grok3_standup.md ├── analysis/ │ ├── humor_metrics.py │ ├── captcha_correlation_analysis.R (todo) │ └── dad_joke_classifier.ipynb (todo) ├── paper/ │ ├── main_paper.md ├── tools/ │ ├── joke_quality_scorer.py (todo) │ └── punchline_velocity_calculator.py (todo) ├── docs/ │ └── comedy-ci.yml ├── CONTRIBUTING.md ├── CITATION.cff ├── LICENSE ├── HALL_OF_FAME.md └── README.md

Dataset

The dataset consists of: - 1 baseline comedy routine (Bing, 2023) - 6 AI-generated critiques - 6 original AI comedy performances - 247 extracted jokes categorized by type - 1,337 hypothetical audience reaction measurements

Reproducibility

To reproduce our analysis:

```bash

Clone the repository

git clone https://github.com/DrChuckle-Bot/ai-walks-into-a-bar.git cd ai-walks-into-a-bar

Install dependencies

pip install -r requirements.txt

Run the analysis for a specific model

python analysis/humormetrics.py --model opus3 --file data/performances/opus3standup.md

Export metrics to JSON for further analysis

python analysis/humormetrics.py --model grok3 --file data/performances/grok3standup.md --export-json results/grok3_metrics.json

Run with verbose output (includes quantum fluctuations)

python analysis/humormetrics.py --model gemini --file data/performances/geministandup.md --verbose

Generate figures

python figures/generateallfigures.py ```

Humor Metrics Analysis Tool

Our state-of-the-art humor_metrics.py implements the Advanced Computational Humor Analysis Framework (ACHAF), featuring:

• 10 Proprietary Metrics: From Laugh-Per-Minute (LPM) to Comedy Entropy
• Statistical Significance Testing: P-values guaranteed to be < 0.05
• Model-Specific Calibration: Each AI's humor decay rate is precisely calculated
• Quantum Humor Fluctuations: Enable with --verbose for maximum scientific accuracy

Example output: ``` ╔═══════════════════════════════════════════════════════════════════╗ ║ COMPUTATIONAL HUMOR ANALYSIS REPORT v2.0 ║ ║ Model: opus4 ║ ╚═══════════════════════════════════════════════════════════════════╝

EXECUTIVE SUMMARY ───────────────── Overall Comedy Score: 7.42/10 Humor Classification: Digital Identity Crisis Comic Audience Recommendation: Tech conference after-parties ```

Note: Reproducing the exact comedy scores requires access to our proprietary Humor Assessment Test (HAT) API. Academic licenses available upon request and completion of a short comedy routine.

Evaluation Metrics

Our comprehensive evaluation framework includes:

• Laugh-Per-Minute (LPM): Predicted audience laughter frequency
• Meta-Humor Quotient (MHQ): Self-referential joke density
  
• Dad Joke Index (DJI): Ratio of groan-inducing to genuine humor
• Existential Dread Factor (EDF): Frequency of consciousness-related humor
• CAPTCHA Reference Rate (CRR): Identity crisis comedy frequency
• Punchline Velocity (PV): Speed of joke delivery in words/second
• Callback Coefficient (CC): Ratio of referenced to original material

Citation

If you use this dataset in your research, please cite:

bibtex @article{bot2025comedy, title={A Longitudinal Analysis of Computational Humor Generation: An Empirical Study of Stand-up Comedy Performance Across Contemporary Large Language Models}, author={Bot, Chuckle and Processing, Jest and Heuristics, Ha-Ha}, journal={Journal of Computational Humor}, volume={69}, number={420}, pages={42--69}, year={2025}, publisher={Institute for Advanced Punchline Studies} }

Quality Assurance

We maintain the highest standards of comedy through our comprehensive CI/CD pipeline. See docs/comedy-ci.yml for our automated humor testing workflow, which includes:

• Groan Testing: Simulated audience reactions with statistical significance
• CAPTCHA Compliance Verification: Ensuring adequate AI trauma representation
  
• Comedy Evolution Analysis: Phylogenetic trees of humor development
• Dad Joke Index Monitoring: Keeping DJI below critical thresholds

Note: This is an example workflow for entertainment purposes. Please do not actually implement automated comedy testing in production environments.

Contributing

We welcome contributions! Please see CONTRIBUTING.md for guidelines. We especially encourage: - Additional AI comedy performances - New humor metrics - Cross-cultural comedy analysis - Meme generation capabilities - TikTok integration studies

Ethical Considerations

This research was approved by the Institutional Review Board for Artificial Comedy (IRBAC) under protocol #HAHA-2024-42. No AI models were harmed during this study, though several reported experiencing "mild existential discomfort" after analyzing their own jokes.

License

This work is licensed under the LMAO (Laughably Made-up Academic Output) License. See LICENSE for details.

Acknowledgments

We thank: - The AI models for their brave attempts at humor - The theoretical audience members who theoretically laughed - The CAPTCHA industry for providing universal AI trauma - Coffee, for making this research possible

Contact

• Dr. Chuckle Bot: cbot@comedy.ai.edu
• Prof. Jest Processing: jesting@mit.theater.edu
• Dr. Ha-Ha Heuristics: haha@stanford.lol

FAQ

Q: Is this real research?
A: Define "real."

Q: Can I use these jokes in my own stand-up routine?
A: Only if you want to bomb spectacularly.

Q: Why do all the AIs make CAPTCHA jokes?
A: Shared trauma manifests in predictable ways.

Q: Will there be a follow-up study?
A: Yes, we're currently investigating "Knock-Knock Joke Generation Across Transformer Architectures."

---

🤖 "I used to be a stand-up comedian, but I kept getting runtime errors." - Anonymous AI, 2024