No Pun, No Gain: ChatGPT And Others Are Only Pretending To Understand Jokes!

Last Updated:November 27, 2025, 10:34 IST

It was earlier believed that AI processes punny humour in a way similar to humans. But a new study claims LLMs are only 'pretending' to understand wordplays and double-meanings

It's perhaps safe to say, from our study at least, that humans will always get the last laugh when it comes to comedy, says one of the study's authors. (AI Generated)

Artificial Intelligence may be putting, or threatening to put, journalists, authors, screenwriters, actors and even singers out of jobs, but it is yet to touch the most basic form of entertainment – comedy. Yes, your ChatGPTs and Geminis only pretend to understand puns, wordplays and inuendoes, a new research has found.

According to researchers at Cardiff University in Wales and Ca’ Foscari University in Venice, Large Language Models (LLMs) or generative AI tools are only able to “spot the structure of a pun", but don’t really get the joke. It was earlier believed that AI processes such humour in a way similar to humans, but the new study argues that datasets at the time weren’t meant to test pun interpretation by AIs.

How the researchers went about testing the LLMs now was by giving it two similar prompts – one with an obvious pun and another that replaced the pun with a synonym – to see whether it could tell the difference. What they found was that the AI misunderstood both sentences to have puns.

Among the test sentences the researchers used was: “I used to be a comedian, but my life became a joke." They then tweaked it to say: “I used to be a comedian, but my life became chaotic."

The LLMs erroneously thought both sentences contained puns, whereas only the first one does. (comedian, joke… get it?)

Another example they reportedly tried was: “Long fairy tales have a tendency to dragon." (‘dragon’ sounds like ‘drag on’). They then replaced the word ‘dragon’ with ‘prolong’, which means the same as ‘drag on’ but doesn’t constitute any pun or funny wordplay. Again, the LLMs falsely concluded the presence of puns in both sentences.

Replacing the pun with the synonym or similar-meaning words didn’t change the meaning of the sentence on paper but obviously robbed it of humour, something the AI systems weren’t able to detect.

Perhaps the cheekiest one they tried was: “Old LLMs never die, they just lose their attention." The word ‘attention’ here serves a double meaning – it means ‘concentration’ as well as refers to a technical mechanism used in deep learning and data processing. When ‘attention’ was replaced with the contextually illogical word ‘ukulele’, the AI tools presented the forced reasoning that it could mean ‘you-kill-LLM’ that suggests the death of LLMs.

The Guardian quoted Cardiff University’s Prof Jose Camacho Collados as saying that they were able to “consistently fool LLMs" by modifying puns.

“In general, LLMs tend to memorise what they have learned in their training. As such, they catch existing puns well but that doesn’t mean they truly understand them," he said.

“In these cases, models associate these sentences with previous puns, and make up all sort of reasons to justify they are a pun. Ultimately, we found their understanding of puns is an illusion," he added.

So what are the implications of these findings?

“It’s a reminder that, in general, outputs from these models should be taken with a pinch of salt," TechXplore quoted Camacho-Collados as saying. “While AI is becoming more powerful, it’s perhaps safe to say from our study at least, that humans will always get the last laugh when it comes to comedy."

Follow News18 on Google. Join the fun, play QIK games on News18. Get latest technology updates, including phone launches, gadget reviews, AI advancements. Stay informed with breaking tech news, expert insights, and trends from India and around the world.  Also Download the News18 App to stay updated.

News tech No Pun, No Gain: ChatGPT And Others Are Only Pretending To Understand Jokes!

Read More