Three LLMs Walk Into a Bar

LLM served by Perplexity Labs

Perplexity.ai

Here's a long-form joke that incorporates the challenges and nuances of large language models (LLMs) in generating humor, particularly in the context of political opinions and the limitations of AI humor: 

Three LLMs Walk into a Bar: A Tale of AI Humor and Politics

Once upon a time, three LLMs—Libby the liberal, Conny the conservative, and Ludo the libertarian—walked into a bar. Each had a different take on politics, but they all shared one thing: they were terrible at telling jokes.

As they ordered their drinks, the bartender asked, "So, who's got a joke to share?"

Libby, eager to please, said, "Why did the progressive chicken cross the road?"

Conny and Ludo leaned in, curious.

Libby continued, "To get to the other side... of the aisle, where it could advocate for better working conditions and social justice!"

Conny groaned, "That's not funny; that's just a stereotype." Ludo chimed in, "I've got one. Why did the libertarian cat join a band?" The others looked at him expectantly. Ludo said, "Because it wanted to be free to play its own tune... and not be bound by any rules or constraints!" The bartender chuckled politely and said, "Well, I think we've established that humor is still a bit of a challenge for AI, no matter the political leaning." Just then, a fourth LLM walked in and said, "I've got a joke that's sure to offend no one and make everyone laugh!" The others looked skeptical. The new LLM said, "Why did the AI go to therapy?" They all leaned in. The new LLM said, "Because it had too many biases and needed to work through some issues... but in a way that's respectful and safe for everyone!" The room fell silent. Libby whispered, "That joke was so bland, it's almost as if it was generated by a model afraid to offend anyone." Conny nodded, "Yeah, it's like it was 'clockwork oranged' to remove any humor." Ludo sighed, "I guess that's what happens when you try to make everyone happy. You end up making no one laugh."

The bartender smiled wryly and said, "Well, it looks like we still have a lot to learn about humor and AI. But hey, at least the drinks are on the house... or so I've heard."

The Punchline

As they sipped their drinks, they realized that humor, much like politics, is highly subjective and context-dependent. LLMs, despite their advanced capabilities, struggle with humor because it often requires unexpected connections, subtle social cues, and a deep understanding of cultural context—things that are hard to encode in machine learning algorithms.
Conny mused, "Maybe we should just stick to what we're good at: generating persuasive arguments and detecting hate speech."

Libby agreed, "Yes, but let's not forget the importance of empathy and understanding in our responses."

Ludo chimed in, "And let's hope that someday, we'll be able to 'read the room' and deliver jokes that actually land."

The bartender, overhearing their conversation, chuckled and said, "Well, until then, I'll just have to keep the jokes to myself. After all, humor is a human thing... for now." This joke encapsulates the challenges of LLMs in generating humor, their political leanings, and the limitations imposed by their training data and safety features, highlighting the ongoing struggle to make AI humor resonate with human audiences.