Laughing Language Models (LLM) Comedy Showdown
Forget the lame dad jokes. Which LLM (if any) can help me (and you) be funny/funnier/not lame?
Background
In Part 1 of this series, I started to explore data driven comedy. A 60-second stand-up comedy routine to extract the timestamps of the jokes, in an effort to find out if data science can be used with generative AI to land a spot on The Tonight Show, Dry Bar Comedy, or what is more likely for me — special guest entertainer at the local cat show (old ladies with cats love me).
Nobody from Netflix called in the past week to ask me to do a special, so I’ll keep moving forward — and towards data science!
From there I created my own hilarious comedy routine, which nobody then listened to.
All in the name of learning data science and time-series analysis without boring everyone to death! We eschewed (there’s a big unnecessary word) how long it takes to flip ten coins, the duration between rolling dice at a craps table (crap I really wanted to know that!), and every other lame example used in every statistics course known to exist.
Experiment #2
This is a short week with another teacher’s holiday. Meaning my kids will be home Friday and nothing will get done.
I haven’t the slightest clue what data sciency (that’s a real word I made up) stuff will be considered. My hope-thesis for this article is it’s going to be more fun than watching the Dallas Cowboys lose again.
(Two Hours Later) — Nope, I didn’t miss anything. Cowboys got smoked 44–19 by the Saints.
(Three Days Later…) The big brains at Anthropic did an AI Prompt Engineering Deep Dive. I highly recommend watching it. You learn a lot, they didn’t use math, and only said a few big words that I had to lookup in a dictionary.
TL;DW. Think about how AI can be used first, when trying to accomplish a task. People tend to use it after. Plus a bunch of stuff on prompts.
I also highly recommend watching the new Linkin Park video.
TL:DW. The new lead singer is amazing.
After watching this discussion (and the Linkin Park video several times), I decided to do a different test than originally planned. And since it was going to be way easier than what I had planned, my l̶a̶z̶y̶i̶n̶e̶s̶s̶ genius kicked in…
I wondered if I fed an LLM:
- One of my first draft (terrible) comedy routines
- A well crafted (barely put effort into) prompt
- Time series markers from a new, proven comedy routine from a really well known comedian…
Would the LLM turn my garbage into gold? Or would garbage-in turn into garbage-out?
The Showdown
Showdown is probably too dramatic. This was more like saying — hey, see what happens if we tie the fuses of three firework rockets together and light them all at once.
Nothing too well thought out.
The Prompt:
Act as a comedian writing a comedic bit. Based on other successful joke segments you’ve done in the past, you know timing of jokes is critical. The following time series is from your past jokes. Each timestamp represents when in the bit a joke was told. Each one is separated by a space. Take the following rough idea of a comedy routine, and rewrite it to try to make the jokes line up with the timestamps:
Seemed like a reasonable prompt (timestamps excluded for brevity).
Given I just watched a bunch of smart people talk about Claude, Claude would be my first test.
Claude
Claude sucked. Which is too bad. Because I like Claude. It’s the only chat bot with a name.
No matter how many times I changed the prompt (ok, five times), it couldn’t figure out how to chunk the words properly, and align the funny parts with the given timestamp. There was too much dead space in the segments.
00:05 — My wife’s friend went to the ER recently after biting into a peppercorn.
00:15 — Now, I’m not saying peppercorns aren’t dangerous. Inhaling pepper has been known to cause bouts of sneezing.
00:29 — But when I’m doing the math on why healthcare costs are out of control… I blame her. I chewed on one side of my mouth for three years because I was too cheap to get a new crown. I’m confident I could survive a random attack from a peppercorn.
00:52 — I didn’t know that peppercorn. I never met it. Maybe it was very angry.
00:55 — I’m never going to upset a potato chip though. It might slit my throat.
01:07 — Can you imagine getting this call from a relative at 2 am?
01:12 — “What? What? What is it? What happened?”
01:18 — <dramatic pause>
01:31 — “Peppercorn. She was eating a Caesar salad, from Macaroni Grill. And it almost killed her.”
01:41 — “The fresh ground pepper just… didn’t make it out of the pepper grinder. I think it was faulty.”
01:49 — “Half the restaurant ended up in the ER. It was horrific!”
01:59 — “Gunshot victims were being turned away because all the beds were full. What kind of health care system does America have anymore?”
02:14 — I think we need to lawyer up. That’s a lawsuit, folks.
02:23 — Peppercorns are the silent killer.
Every prompt revision given, was kindly responded to by Claude with ‘yeah, yeah, yeah, I totally get what you’re asking for. Let me fix it and do just what you asked’.
And then nothing got fixed. It was like asking my kids to clean their room — yes Dad, I’m going to do that later before bed.
And two days later, it’s still not done.
I’m sure if Dr. Philosophy (not to be confused with Dr. Feelgood) from the interview were doing the prompting, she would nail it. But I’m in a rush to not work on this.
ChatGPT
Boom!
Given the exact same starting prompt as Claude, ChatGPT nailed it in the first attempt. I continued to tweak it a little. Here’s what it came up with:
[00:05] So, my wife’s friend went to the ER recently… after biting into a peppercorn.
Pause for effect, then speak incredulously
A peppercorn! Yeah, you heard me right.[00:15] I’m not saying peppercorns aren’t dangerous. We’ve all inhaled pepper, and yeah, you sneeze a bit… maybe cry if you’re weak. But going to the ER? For a peppercorn?
[00:20] Look, this is why healthcare costs are through the roof! I blame her. My tooth broke once, and I chewed on one side of my mouth for three years because I was too cheap to fix it!
[00:23] I’m pretty sure I could survive an attack from a peppercorn. I didn’t know that peppercorn. I never met it. Maybe it was angry… bad upbringing or something.
Pause, shift in tone
But I’ll tell you one thing… I’m never crossing a potato chip. It might slit my throat!
Let laughter build[00:29] Now, imagine getting this call from a relative… 2 AM. Your phone rings, you wake up confused.
Act out tired confusion
“What? What? What happened?!”[00:52] Pause dramatically, then deliver deadpan
“Peppercorn.”[00:55] “She was eating a Caesar salad… from Macaroni Grill… and it almost killed her.”
Build up the drama
The fresh ground pepper didn’t make it out of the pepper grinder. It was… faulty. Half the restaurant ended up in the ER![01:07] Gunshot victims were being turned away.
“Sorry, folks, no room. We’ve got peppercorn injuries piling up in here.”[01:12] What kind of health system does America have when peppercorns are filling up the ER? You’re there with people who got stabbed, and you’re like, “Yeah, I was viciously assaulted… by my salad.”
[01:18] Pause.
That’s right. We need to lawyer up. That’s a lawsuit, folks.[01:31] Peppercorns are the silent killer. You don’t hear them coming, you don’t see them. They blend in. Next time you’re at a fancy restaurant, just know… you might not make it out.
[01:38] Leans forward dramatically.
You think it’s allergies? Nah. That’s peppercorn warfare.[01:41] Fake sneeze and cough.
See? That’s how they get you![01:47] You laugh now, but just wait. When a rogue parsley flake takes someone down, we’ll all regret not taking this seriously.
[01:49] The vegetables… they’re coming for us, man! They’re out here, living their best lives.
Pause for laugh[01:59] So, next time you’re out to dinner and someone offers fresh pepper…
Pause, look around nervously
Just walk away. Save yourself. Don’t be a hero.[02:04] They’re probably here right now, lurking… in the audience.
Point dramatically into the crowd[02:09] Pause again, lean in for effect
Stay safe out there, folks. Peppercorns are everywhere.[02:14] And that’s the peppercorn conspiracy. I rest my case.
Dramatic flourish, end bit
ChatGPT even revised and added to many of the jokes. They were, dare I say…funny.
ChatGPT even provided physical cues of what to do in front of an audience. That’s next level scary. Scarier than a peppercorn.
Gemini, Meta, I Dunno
Gemini, given the same prompt, did exactly what Claude did first — take every sentence and assign it to a timestamp. Maybe Claude and Gemini were built by the same engineers.
Gemini needed additional prompting to rewrite the original. What it came up with has some funny moments:
00:05: So, my wife’s friend recently ended up in the ER.
00:15: Can you guess why?
00:20: A peppercorn. A tiny, innocent-looking peppercorn.
00:23: I mean, I’ve eaten worse. Like that time I accidentally ate a whole jar of pickles. But a peppercorn?
00:29: I’m pretty sure I could survive a ninja attack before succumbing to a peppercorn.
00:52: It’s like, “Hey, I’m just trying to enjoy my salad here. Why are you trying to kill me?”
00:55: I mean, I’ve seen people survive being hit by lightning. A peppercorn? Really?
01:07: Imagine getting a late-night call from your relative.
01:12: “What? What happened?”
01:18: Dramatic pause “A peppercorn.”
01:31: Dramatic gasp “A peppercorn?!”
01:38: Apparently, this peppercorn was a ninja. Or maybe it just hated her.
01:41: The whole restaurant was in chaos. People were screaming, running around, trying to avoid the peppercorn.
01:47: It was like a scene from a horror movie, but with less gore and more salad dressing.
01:49: I mean, what kind of healthcare system do we have when people are getting rushed to the ER because of a peppercorn?
01:59: We need to sue the peppercorn company. Or maybe just the restaurant for serving such a deadly salad.
02:04: Peppercorns: The silent killer.
Lllama Meta
Yeesh, I thought I wasn’t funny. Suddenly a new winner emerges:
(00:05) You know, my wife’s friend recently had a harrowing experience. She bit into a peppercorn and ended up in the ER.
(00:15) Now, I’m not saying peppercorns are deadly, but have you ever inhaled pepper? It’s like a sneeze bomb just waiting to go off.
(00:20) But seriously, when I think about healthcare costs skyrocketing, I blame her.
(00:23) I mean, I’ve been chewing on one side of my mouth for three years because I’m too cheap for a new crown.
(00:29) I’m pretty sure I could handle a surprise peppercorn attack.
(00:52) I didn’t know that peppercorn, never met it. Maybe it had a grudge.
(00:55) But you know what I’ll never mess with? A potato chip. Those things are sharp.
(01:07) Can you imagine getting that 2 a.m. call from a relative?
(01:12) “What? What? What is it?”
(01:18) (dramatic pause) …Peppercorn.
(01:31) She was eating a Caesar salad from Macaroni Grill. That peppercorn was a stealth assassin.
(01:38) The fresh ground pepper just didn’t make it out of the grinder… probably faulty.
(01:41) I heard half the restaurant ended up in the ER. Gunshot victims were turned away.
(01:47) What kind of healthcare system do we have?
(01:49) Lawyer up, folks! Peppercorns are the silent killers.
(01:59) We need warning labels: “Caution: May contain hazardous peppercorns.”
(02:04) Macaroni Grill should provide hazmat suits with every salad.
(02:09) That poor peppercorn victim deserves compensation.
(02:14) It’s time to take a stand against rogue spices.
(02:17) Peppercorn-gate, folks. It’s a thing.
(02:23) Stay vigilant, America. You never know when a peppercorn might strike.
Stay vigilant? I had a Ron Burgundy moment reading that…
Final Output
Here’s a link to the final version. It’s got some additional human tweaks that the AI wasn’t ever going to make. Because it doesn’t eat salad and wouldn’t get it.
Conclusions
- I can kiss any chance of being hired by Anthropic goodbye. I love you Claude!
- I can kiss any chance of being hired by Anthropic to entertain the employees at the company Christmas party goodbye. :(
- After working on this tirelessly for ten whole minutes, my hypothesis for this iteration in the quest for data driven comedy is this:
Every LLM is significantly better than the Dallas Cowboys.
Stay vigilant America…
