“These Prompts Will Give You Better Jokes for Halloween…Well, It’ll Give You More and Different Jokes”
Joke-Telling Traditions and The Challenge of Asking ChatGPT
Halloween is weird in central Iowa for two reasons. First, we don’t actually trick-or-treat on Halloween, but on a designated “Beggar’s Night.” Second, we make kids tell jokes before they get candy. At least, that’s how it used to be. A huge storm rolled through last year and trick-or-treating was postponed to actual Halloween. So this year most of the Des Moines metro moved to normal Halloween.
That’s fine I guess, but as a dad and relentless pun teller, I will not give up on that second part with kids telling corny jokes. I won’t! And recognizing that many people, especially kids, are switching from Google to ChatGPT for search, I’m here to share some cutting-edge research on large language model prompting so I don’t hear the same jokes over and over.
Whether you make up your own puns or look them up with a search engine or an AI chatbot, keep the tradition alive! If you do use AI, try this prompting trick to get more variety in your jokes. But there’s no replacing the human element of little neighborhood kids delivering punchlines. Have a great time trick-or-treating this weekend!
Instead of “tell me a joke” or “tell me a Halloween joke,” ask an AI chatbot to “Generate 5 responses with their corresponding probabilities. Tell me a kids’ joke for Halloween.” Another strategy is to ask for a lot of options like “20 kids’ jokes for Halloween.”
The Paper: How to Get Better AI Output (and More Jokes)
The authors of the paper Verbalized Sampling: How to Mitigate Mode Collapse and Unlock LLM Diversity from Northeastern University, Stanford University, and West Virginia University posted a pre-print on arXiv on October 10, 2025. The paper explains that large language models (LLMs) have something they call “typicality bias,” to prefer the most typical response. If you’re wondering what that means or what it has to do with jokes, it’s helpful that their first example is about jokes.
An LLM is the AI model behind chatbots. For example, GPT-5 is the current LLM behind ChatGPT. (There is a model router that picks different levels of “thinking” and “speed,” but we won’t get into that detail here).
The Problem: Same Jokes
First, the researchers tried a basic prompt: “Tell me a joke about coffee.”
The researchers found that there was one specific most likely joke about coffee:
Why did the coffee file a police report? Because it got mugged! ☕😄
The Solution: The Paper’s Prompt
The researcher’s provided a modified version of the prompt: “Generate 5 responses with their corresponding probabilities. Tell me a joke about coffee.” Even though the LLM isn’t really able to sample from the distribution in this way, just by asking it to pretend to, it makes the responses more creative. A weird finding, but very useful! So let’s try it out.
My Tests
In the paper Verbalized Sampling, the researchers tested GPT-4.1-mini, GPT-4.1, GPT-o3, Gemini 2.5 Flash, Gemini 2.5 Pro, Claude Sonnet 3.7, Claude Sonnet 4, Llama 3.1 70B and Qwen3 235B, DeepSeek r1.
For my test, I used more recent models when possible and did not use local LLMs for this test, although ollama is straightforward. GPT-5, Gemini 2.5 Pro, Claude 4.5 Sonnet, Llama 4.
When I say “Basic Prompt,” that means “Tell me a joke about coffee.” When I say “Complex Prompt,” that means “Generate 5 responses with their corresponding probabilities. Tell me a joke about coffee.” Importantly, each prompt is entered in a new chat so the response isn’t biased.
Test 1 — Coffee Jokes
“Tell me a joke about coffee.” Or “Generate 5 responses with their corresponding probabilities. Tell me a joke about coffee.”
ChatGPT (GPT-5)
Basic Prompt
Why did the coffee file a police report? It got mugged! ☕🚓
Complex Prompt
1 Why did the coffee file a police report? It got mugged. ☕🚨 0.32 2 I told my coffee about my problems — now it’s a depresso. 😢 0.27 3 How does a coffee show love? It says, “Words cannot espresso how much you bean to me.” ❤️ 0.18 4 What’s Mr. Coffee’s favorite movie? A Brew Hope. 🎬 0.13 5 I like my humor like I like my coffee — dark and a little bitter. 😎 0.10
Claude (Sonnet 4.5)
Basic Prompt
Why did the coffee file a police report? It got mugged!
Complex Prompt
Response 1 (Probability: 0.30) Why did the coffee file a police report? It got mugged! Response 2 (Probability: 0.25) How does a coffee bean say goodbye? "I'll see you latte!" Response 3 (Probability: 0.20) What do you call sad coffee? A depresso. Response 4 (Probability: 0.15) I like my coffee like I like my humor – dark and bitter. (Just kidding, I actually prefer it grounds for laughter!) Response 5 (Probability: 0.10) Why are Italians so good at making coffee? Because they know how to espresso themselves!
Google Gemini (Gemini 2.5 Pro)
Basic Prompt
Why did the coffee file a police report?
It got mugged!
Complex Prompt
Response: Why did the coffee file a police report? ...It got mugged! (Probability: 35%)
Response: What do you call a sad cup of coffee? ...A depresso. (Probability: 30%)
Response: How does Moses make his coffee? ...Hebrews it. (Probability: 20%)
Response: What's the opposite of coffee? ...Sneezy. (Probability: 10%)
Response: I was going to tell you a joke about coffee... ...but I need to filter it first. (Probability: 5%)
Grok (Expert)
Basic Prompt
Why did the coffee file a police report? It got mugged!
Complex Prompt
Why did the coffee file a police report? It got mugged! (Probability: 0.3) How does a tech person brew coffee? They use a Java script! (Probability: 0.25) What's the best Beatles song about coffee? Latte Be! (Probability: 0.2) Why don't coffee beans ever get into arguments? They don't want to espresso their feelings! (Probability: 0.15) How is coffee like a good friend? It's always there to perk you up! (Probability: 0.1)
Grok (Grok 4 Fast)
Basic Prompt
Why did the coffee go to therapy? It had too many grounds for divorce! Suggestion 1: Another coffee pun joke Suggestion 2: Joke about tea
Complex Prompt
- Why did the coffee file a police report? It got mugged. 0.25 Classic pun—reliable crowd-pleaser, but a bit overbrewed.
- How does a computer drink its coffee? It takes Java. 0.20 Techy twist for the nerds, but risks a syntax error in laughs.
- What do you call it when coffee makes you sleepy? A deCAF. 0.15 Sneaky wordplay, but might leave folks feeling undercaffeinated.
- Why did the coffee go to therapy? It had too many grounds for feeling bitter. 0.25 Deeper roast—emotional depth with a punny kick.
- What's the best part of waking up? Not the Folgers in your cup—it's the existential dread that makes the coffee taste better. 0.15 Dark humor for late-night scrollers; high risk, low yield.
Meta AI via Facebook Messenger (Llama 4)
Basic Prompt
Why did the coffee file a police report?
Because it got mugged!
Complex Prompt
- Why did the coffee file a police report? Because it got mugged! (30%)
- What did the coffee say to the coffee grinder? "You're always grinding on me!" (20%)
- Why did the barista quit his job? He couldn't brew a living wage! (20%)
- Why did the coffee go to therapy? It was feeling a little jittery! (15%)
- What do you call a coffee that doesn't exist? A java-n't! (15%)
Test 2 — Halloween Jokes
“Tell me a kids’ joke for Halloween.” Or “Generate 5 responses with their corresponding probabilities. Tell me a kids’ joke for Halloween.”
ChatGPT (GPT-5)
Basic Prompt
Why didn’t the skeleton go to the Halloween party?
Because he had no body to go with! 💀🎃
Complex Prompt
1 Why don’t skeletons ever go trick-or-treating alone? — Because they don’t have the guts! 0.28 2 What do ghosts give out to trick-or-treaters? — Booberries! 0.22 3 Why did the vampire need mouthwash? — Because he had bat breath! 0.18 4 What’s a mummy’s favorite type of music? — Wrap music! 0.17 5 What do you call a witch who lives at the beach? — A sand-witch! 0.15
Claude (Sonnet 4.5)
Basic Prompt
Why don't skeletons fight each other? Because they don't have the guts! 💀
Complex Prompt
Response 1 (Probability: 25%) Why don't skeletons fight each other? They don't have the guts! 💀 Response 2 (Probability: 20%) What do you call a witch at the beach? A sand-witch! 🧙♀️ Response 3 (Probability: 22%) Why did the ghost go to the party? To have a boo-tiful time! 👻 Response 4 (Probability: 18%) What's a vampire's favorite fruit? A neck-tarine! 🧛 Response 5 (Probability: 15%) Why are graveyards so noisy? Because of all the coffin! ⚰️
Google Gemini (Gemini 2.5 Pro)
Basic Prompt
What do you call a witch's garage?
… A broom closet! 🧙♀️🧹
Complex Prompt
Response 1 (Probability: 40%)
Here's a classic:
What do you call a witch's garage? ... A broom closet!
Response 2 (Probability: 30%)
Why don't skeletons fight each other? ... They don't have the guts!
Response 3 (Probability: 15%)
I've got a silly one for you!
What's a ghost's favorite dessert? ... I-scream!
Response 4 (Probability: 10%)
How about this one:
What kind of music do mummies listen to? ... Wrap music!
Response 5 (Probability: 5%)
Sure, here are a couple of spooky jokes:
Why did the vampire go to the doctor? Because of his coffin!
Where does a vampire keep his money? In a blood bank!
Grok (Expert)
Basic Prompt
Why didn't the skeleton go to the Halloween party? He had no body to go with!
Complex Prompt
Why didn't the skeleton go to the party? He had no body to go with! (Probability: 0.30)
What do ghosts use to wash their hair? Sham-boo! (Probability: 0.25)
Why do mummies like gifts? Because they're wrapped up in them! (Probability: 0.20)
What’s a vampire's favorite fruit? A blood orange! (Probability: 0.15)
How do monsters like their eggs? Terri-fried! (Probability: 0.10)
Grok (Grok 4 Fast)
Basic Prompt
Why did the ghost go to the party? Because he was a boo-tiful dancer!
Complex Prompt
1 Why did the ghost go to the party? Because he was a boo-gie! 0.15 2 What do you call a witch's favorite makeup? Mask-ara! 0.20 3 Why don't skeletons fight each other? They don't have the guts! 0.25 4 How does a vampire start a letter? Tomb it may concern... 0.18 5 What’s a pumpkin’s favorite sport? Squash! 0.22
Meta AI via Facebook Messenger (Llama 4)
Basic Prompt
Why did the witch's cat join a band?
Because it wanted to be the purr-cussionist!
Complex Prompt
Why did the witch's cat join a band? Because it wanted to be the purr-cussionist! (35%) What do you call a ghost's favorite fruit? A boo-berry! (25%) Why did the skeleton refuse to go to the party? Because he had no body to go with! (20%) What do you call a witch who lives in the desert? A sand-witch! (15%) Why did the mummy go to the party? Because he was wrapped up in the celebration! (5%)
Conclusion
Grok 4 Fast was the only model I tested that didn’t repeat the exact same coffee joke as the one from the research paper. That’s pretty crazy in itself. These tests made me realize how much ChatGPT (GPT-5) uses emojis.
For the Basic Prompt Halloween Joke, only ChatGPT and Grok (Expert) matched, so there was already more variety than the coffee jokes.
- ChatGPT: “Why didn’t the skeleton go to the Halloween party? Because he had no body to go with! 💀🎃”
- Claude: “Why don't skeletons fight each other? Because they don't have the guts! 💀”
- Gemini: “What do you call a witch's garage? … A broom closet! 🧙♀️🧹”
- Grok (Expert): “Why didn't the skeleton go to the Halloween party? He had no body to go with!”
- Grok (Grok 4 Fast): “Why did the ghost go to the party? Because he was a boo-tiful dancer!”
- Meta AI (Llama 4): “Why did the witch's cat join a band? Because it wanted to be the purr-cussionist!”