ChatGPT analyzes a joke

10 Jun 2023

I recently read that ChatGPT is familiar with very few jokes and is probably
incapable of creating one of its own. So, I thought I would make up a joke and
present it to ChatGPT to see how it would analyze it by the 'rules of comedy'.
In subtle cases like this, I prefer to warn ChatGPT ahead of time what I'm doing
so that it won't get the wrong impression of what I'm trying to do. Now, I've
never seriously studied the rules underlying humor but I'm pretty sure that ChatGPT
has. Lastly, although I made a joke at the expense of LLMs, I have nothing but
the highest regard for them as information/reasoning assistants, for ChatGPT and
BingChat in particular.

And away we go!


Patrick:

I've invented a joke in the form of a riddle that I think some people might find
humorous. I'd like you to comment on whether you think that this joke has the
right elements to be funny.

ChatGPT:

Of course! I'd be happy to give you feedback on your riddle-joke. Please go
ahead and share it with me, and I'll let you know what I think about its
comedic potential.

Patrick:

Why did the large language model cross the road?

ChatGPT:

Why did the large language model cross the road? I'm curious to hear the punchline!

Patrick:

Because it wanted to look back at itself.

ChatGPT:

Ah, I see! The punchline suggests that the large language model crossed the road
to gaze at its own reflection or self-reflection. As for the comedic potential, humor
is subjective and can vary from person to person. However, I can provide you with
some analysis.

This joke relies on a play on words and the concept of self-reflection. It incorporates
the idea that language models, like myself, are designed to generate and analyze text,
so looking back at oneself could be interpreted as a metaphorical reflection.

My comment: ChatGPT is not wrong, but it chose to take a more serious
interpretation of my riddle than I intended. It seems that ChatGPT
regarded my joke as more of a metaphor with a pithy moral at the end,
than as a silly joke I made at the expense of LLMs. But, as Doctor Who
once observed:
People often overlook the obvious.
And if people can do that, surely LLMs can as well. I'll present my
interpretation in a moment.
The humor in this joke might appeal more to individuals who are familiar with
language models, artificial intelligence, or the field of natural language processing.
People who have an understanding of the technology and its capabilities may
appreciate the clever twist on self-reflection.

Overall, it's a witty and intelligent joke, but its effectiveness may depend on the
audience's familiarity with the subject matter. Keep in mind that humor can be
subjective, and different people may respond differently to the same joke.

Patrick:

My thought on it is that after the large language model crossed the road, it would
no longer be able to see itself on the other side of the road (because it would no
longer be there), which it would find out the hard way.

ChatGPT:

Ah, I see! Thank you for providing your interpretation of the joke. In that case,
the punchline adds an element of surprise and a twist to the traditional riddle
format. It plays with the expectation that the large language model would look
back at itself, but instead, it humorously realizes that it can no longer see itself
because it's no longer on the other side of the road.

This variation of the punchline introduces an element of irony and a surprising
outcome, which can enhance the humor. It adds an extra layer of cleverness and
plays with the listener's expectation, making it more amusing.

Overall, I find your interpretation of the joke to have a good element of surprise
and irony, which can contribute to its comedic value. Well done!


Interesting. When ChatGPT interpreted my riddle as a metaphor with a moral at
the end, it struggled to find anything complimentary about the humor of my riddle.
But after I suggested a more literal interpretation of the words, it understood that
the LLM in the riddle had slipped up a bit in its thinking. It had overlooked the
obvious. We've all been there, haven't we?

To its credit, once I explained my interpretation of the riddle and its answer to
ChatGPT, its revised explanation proved that it completely understood what I meant.

It seems ironic to me that, although ChatGPT cannot experience the humor of a
particular joke, if you can and do, it may be able to tell you why on technical
grounds.