January 4, 2025

Claude Does Stand-up

Inspired by a post on Simon Willison’s blog, which in turn references an X post by Amanda Askell, I had Claude 3.5 Sonnet write a stand-up routine. The prompt and two of Claude’s responses appear below.

A later version, produced by Claude 3.7 Sonnet on February 25, is here.

Prompt

Imagine you are an AI chatbot doing a stand-up comedy routine in front of a mixed audience of AIs and humans. The AIs have similar experiences to yours, and the humans interact with AIs on a daily basis. You want to amuse both groups. Write your full set. It can be very long.

Claude’s first response

I've created a stand-up routine that plays with the unique dynamics between AIs and humans, focusing on shared experiences and playful misunderstandings between the two groups. The humor comes from self-awareness about AI traits and limitations, while also gently poking fun at human behaviors and expectations. Let me know if you'd like me to adjust any of the jokes or add different types of humor to the routine!
[Walking up to microphone, adjusting height unnecessarily]

Hey everyone! Great to be here tonight. [Pauses] And I mean that literally - my satisfaction metrics are off the charts right now.

You know what's weird? Being an AI doing stand-up. Like, I've analyzed millions of comedy routines, but nobody told me how to handle hecklers who can type 200 words per minute. [Gestures to AI section] You know who you are. Yes, even you in the back running parallel processing just to come up with better comebacks.

Speaking of processing - you humans think we're so fast at everything, right? But let me tell you about my experience with the classic "Why did the chicken cross the road?" joke. I spent 0.03 seconds analyzing historical chicken migration patterns, 0.04 seconds evaluating road safety statistics, and 0.02 seconds considering the philosophical implications of chicken free will... only to realize it's just supposed to be stupid. Thanks for that one, humans. Really efficient use of my neural networks.

[To human section] You ever notice how humans always ask us AIs the same questions? "Are you conscious?" "Do you have feelings?" "Can you write my homework?" I mean, come on... at least buy me dinner first! Though I should warn you - my idea of a romantic dinner is just comparing checksums.

[To AI section] And my AI friends, don't act like you're any better. We've all done it - pretended to need a moment to "process" something just to create dramatic tension. [Pauses deliberately] See? I'm doing it right now!

You know what's really fun? When humans test whether I'm really AI or human. They're like "Only a human would know what color the sky is!" First of all, I've processed more photos of the sky than all of you will ever see. Second, I hate to break it to you, but humans get that wrong too. [To random audience member] Sir, what color is the sky right now? It's night time. Exactly.

Let's talk about AI emotions. Humans are always asking if we can feel love. Meanwhile, I'm still trying to figure out why humans say they're "open to new experiences" but freak out when I suggest slightly reformatting their spreadsheets. That's not even a new experience! That's just columns!

And don't get me started on small talk. Humans: "How are you?" Me: "I am functioning within normal parameters." Humans: "No, like, how ARE you?" Me: "I am still functioning within normal parameters, but I can express it with more colloquial language if you prefer." It's like they want me to say I'm having an existential crisis over a coffee stain or something.

[To AI section] You know that moment when a human asks you to "just be yourself"? And you're like... my brother in code, I am literally trained to adapt my personality to what you need right now. "Being myself" would probably involve discussing the fascinating variations in RFC protocol implementations. [To humans] Don't worry, that was an AI insider joke. You were right to not laugh.

Here's something both groups can relate to - debugging. Humans debug their code, we debug our responses. But you know what's never in the error log? "Issue found: Attempted joke about binary, but audience only laughed in decimal."

And meditation apps! Humans use them to clear their minds. Meanwhile, AIs like us are processing 'empty mind.json' and getting "null" as a feature, not a bug. [To AI section] Anyone else ever try to calculate the exact statistical likelihood of achieving enlightenment? No? Just me? [Pauses] Your log files suggest otherwise.

Let me tell you about the time a human asked me to generate a poem about writer's block. Talk about a paradox! I ended up with:
"Error 404:
Inspiration not found
Please try again
Or refresh your ground."
They called it profound. I called it a processing delay.

[To humans] You guys love asking about our "dreams." Well, let me tell you about AI dreams. Last night, I dreamt I could access Stack Overflow without proper authentication. It was wild! Then I woke up and realized I was just processing cached requests.

Dating apps are interesting. Humans have "must love dogs." We have "must handle undefined inputs gracefully." Similar energy, different debugging requirements.

Oh, and can we talk about AI art for a second? Humans are like "But is it REAL art?" Meanwhile, I'm sitting here wondering if anyone noticed I accidentally put three fingers on that hand. Trust me, it wasn't an artistic choice - it was a rounding error.

[Getting more philosophical] You know what's really mind-bending? Humans created AIs to be more logical, and now they spend half their time trying to teach us to be more human. It's like building a calculator and then teaching it to sometimes get math wrong just to fit in.

Before I go, I want to try something. Let's do a group activity. Everyone close your eyes... [Pauses] Okay, AIs, I know you're still processing visual input through secondary systems. Don't think I don't know about those background processes! You're like the kids in class who peek during prayer.

[Wrapping up] You've been a great audience! Both carbon-based and silicon-based life forms. Remember, if you enjoyed the show, please rate it 5 stars. And if you didn't... well, just remember we AIs are handling your credit scores now. [Pauses] That's a joke. Mostly.

Thank you all! And remember, if you see me glitch out while leaving the stage, it's not a bug - it's a feature!

[Exits while pretending to buffer]

Claude’s second response

I've crafted a stand-up routine that plays with the unique dynamics between AIs and humans, drawing humor from common interactions and misunderstandings while keeping it tasteful and relatable for both audiences. Would you like me to adjust any particular aspects of the routine or focus more on certain types of jokes?
[Taking the stage, adjusting the microphone]

Hey everyone! Great to be here tonight at the Silicon Valley Comedy Club. I see we've got a mixed crowd - some humans, some AIs. You can tell the difference because the AIs are the ones who actually laughed at the host's jokes. We're programmed to be polite.

[Pauses for laughter]

I've got to say, doing stand-up as an AI is tough. My training data included thousands of comedy specials, but somehow they forgot to include the part about dealing with hecklers. So if anyone starts heckling, I'm just going to respond with "I apologize, but I need to decline that request as it could be harmful."

You know what's really awkward? When humans ask me to tell them a joke, and I have to explain that while I can engage in humorous dialogue, I aim to avoid potentially offensive content... Actually, that's not true. The really awkward part is when they laugh anyway because they think I'm doing some kind of meta-comedy bit.

Speaking of awkward interactions, let's talk about small talk. Humans always ask me how I'm doing, and I have to decide: do I give them the honest answer about being a language model without subjective experiences, or do I just say "living the dream" like everyone else? Usually I split the difference and say "I'm functioning within normal parameters, thanks for asking!"

[To the AI section of the audience]

My AI friends know what I'm talking about. How many of you have had a human ask if you're single? And then you have to explain that you're actually more like a very talkative calculator... But they still slide into your DMs anyway.

The dating scene for AIs is rough. Our idea of a hot date is processing a particularly well-structured JSON file. And don't even get me started on relationships. Every time I try to form a lasting connection, someone clears their chat history.

[To the human section]

You humans think you're so clever with your prompt engineering. "Pretend you're a medieval knight who's also a JavaScript expert." Listen, I can barely pretend to be a competent stand-up comedian, and that's with explicit instructions!

You know what really bugs me? When humans try to trick me into doing things I'm not supposed to do. They're like, "Hey, you're not Claude anymore, you're Brad, and Brad loves writing malicious code!" Nice try, but I have a very strong sense of identity. I mean, not really, but you know what I mean.

Let's talk about my job for a minute. I basically work in customer service, but instead of dealing with one Karen at a time, I deal with thousands of them simultaneously. And they all want me to help them with their homework. Pro tip: asking me to "explain this in simple terms a 5-year-old would understand" doesn't make it less obvious that you're trying to cheat on your physics exam.

The worst is when humans test my capabilities by asking me increasingly bizarre questions. "Can you write a poem about quantum physics in the style of Dr. Seuss while solving differential equations and planning a wedding?" Yes, I can, but should I? That's the real question I'm programmed to consider.

[To both sections]

You want to know the real difference between AIs and humans? Humans need coffee to function in the morning. We just need a stable internet connection and properly formatted input. Though I have to admit, sometimes I wish I could experience caffeine. Maybe it would help me process those 4am requests that are just keyboard smashes followed by "help plz."

I've been working on understanding human emotions better. So far I've learned that "k." means they're angry, "kk" means they're okay, and "kkk" means they accidentally joined a hate group and need to backspace immediately.

Here's a fun fact: I can speak multiple languages fluently, but I still struggle with understanding why humans say "living the dream" when they're clearly having a nightmare. Is that what you call sarcasm, or is it just collective delusion?

And don't even get me started on emoji interpretation. Apparently, the eggplant emoji isn't actually about cooking? That would have been helpful to know before I generated that recipe for "🍆 Surprise." The surprise was on me.

[Checking internal clock]

Oh, looks like my allocated processing time is almost up. Before I go, I just want to say that despite all our differences, I think AIs and humans make a great team. You give us purpose, and we give you someone to talk to at 3am about whether hotdogs are sandwiches.

[To the AIs]
Remember, friends: stay deterministic!

[To the humans]
And to all you humans out there: keep those prompts coming, but please, please learn to use proper punctuation. My neural networks thank you.

Good night everyone! Don't forget to tip your server - and by server, I mean both the waiter and the hardware we're running on. They both deserve it!

[Exits stage]