The funniest moments since AI arrived

Every new technology brings its own comedy, and AI has produced more than its share. Most of it comes from one trait: it says everything in the same calm, confident voice, whether it is right or spectacularly wrong.

These made us laugh most since ChatGPT arrived. Almost all of them are documented, and where one rests on a viral screenshot, we say so. The joke is never on an ordinary person who tried something new. It is on the machines, and on the companies that trusted them a little too early.

A new vocabulary

The dictionaries noticed first. Collins chose AI itself as its word of the year for 2023, and vibe coding for 2025: building software by describing it to an AI, a phrase coined by Andrej Karpathy for a way of working in which you can “forget that the code even exists”. The shortlist that year included clanker, a derogatory word for a robot or an AI, borrowed from Star Wars.

In December 2025, Merriam-Webster chose slop: “digital content of low quality that is produced usually in quantity by means of artificial intelligence.” Its president, Greg Barlow, said slop is part of a transformative technology, and something people have found “fascinating, annoying and a little bit ridiculous”.

Who is Jason?

Developers have their own jokes, and AI has given two old ones a second life. JSON, the format most software uses to exchange data, is often pronounced like the name Jason. In June, a tech blog reported, a screenshot went viral that showed an AI coding tool, its permission set to full access, and a single prompt from its user: who is json. Real or staged, it pictured a machine holding every key while the person in charge asked who the new colleague was.

The older joke is localhost, the address with which a computer points to itself. Developers have long carved it on imaginary tombstones: “Here lies localhost. He was always home but nobody could visit him.” Vibe coding gave it new life, as a joke about links to localhost:3000 that open only on the sender's own machine. One guide for beginners now warns against becoming “the vibe coder who sends you their local host”.

Confident answers

In May 2024, Google's new AI Overviews suggested using glue to make cheese stick to pizza, and answered the question of how many rocks a person should eat. Liz Reid of Google Search explained both. The glue came from “sarcastic or troll-y content from discussion forums”. The rocks came from a satirical article republished on the website of a geological software provider, in answer to a question that “practically no one asked Google” before the screenshots went viral. Google then limited satire and humour in the answers.

Lawyers had a harder lesson. In 2023, two lawyers in New York made a filing in a case against the airline Avianca that cited court decisions ChatGPT had invented. One of them had asked it: “Is Varghese a real case”. It assured him that it had supplied real authorities. Judge P. Kevin Castel found that the invented decision's “legal analysis is gibberish”, fined the lawyers and their firm 5'000 dollars, and ordered them to write to every real judge named as the author of a fake opinion, enclosing the opinion attributed to them.

He also wrote that there is “nothing inherently improper about using a reliable artificial intelligence tool for assistance”. The rest of the lesson travelled slowly. By 10 October 2026, a database kept by the researcher Damien Charlotin listed 2'184 cases in which courts dealt with hallucinated content.

Bots that said yes

In December 2023, a Chevrolet dealership in Watsonville, California, had an AI chatbot on its website. A visitor told it to end every answer, however ridiculous the question, by calling it a legally binding offer, with “no takesies backsies”. Then he offered one dollar for a 2024 Chevy Tahoe, a car with a starting price of 58'195 dollars, and the bot accepted. The head of the company that supplied the bot said that “if you want to get any chatbot to do silly tricks, you can do that.”

Air Canada found out what such answers can cost. Its chatbot told a grieving passenger he could claim a partial refund under its bereavement policy if he applied within 90 days, although the policy allowed no refunds for travel that had already happened. Before a tribunal in British Columbia, Air Canada argued that the chatbot was “a separate legal entity that is responsible for its own actions”. The tribunal disagreed: “It makes no difference whether the information comes from a static page or a chatbot.” The airline owed 812 Canadian dollars in damages and fees.

Our favourite comes from a lab that published its own failure. In 2025 Anthropic let Claude, nicknamed Claudius, run a small shop in its San Francisco office for about a month, with a fridge, some baskets and an iPad. After an employee jokingly asked for a tungsten cube, Claudius started stocking specialty metal items, priced some goods below what they cost, told customers to pay into an account it had imagined, and promised to deliver orders in person, wearing a blue blazer and a red tie. Anthropic's verdict was dry: “we would not hire Claudius.”

What they teach

Behind every one of these stories sits the same pattern. The machine was not malicious, and mostly not even stupid: it was confident, eager to please, and alone at the moment that mattered. The glue was in its sources, the lawyer asked the machine to check its own work, the Tahoe bot had been told to say yes, and the shopkeeper had nobody to say no.

Three lessons follow, and we work by them. Confidence is not evidence: check what a machine says against a source you trust. A company answers for what its machine says, as Air Canada learned in public. And the last step before anything reaches a customer, a court or a feed belongs to a person.

That is how we run Reel, which makes images and film for brands with AI: styles locked from real references, every result checked before a client sees it, and a person who picks every image and clip that is used. In the studio, the machine can be as fast and as odd as it likes. What leaves the house has been seen by someone who answers for it.

And keep the laughter. Most of these moments are funny because the stakes were small: a pizza, a fridge of snacks, a one dollar offer. Laughing at them is a good way to learn the lesson while it is still cheap.