He makes money trolling AI labs. Now they all want to work with him

=== HEADLINE ===

He makes money trolling AI labs. Now they all want to work with him

A comedian mocks AI models, and now they want to work with him

Simple requests, obvious mistakes

At WIRED World Fair on Wednesday, Husk said that “all the major AI companies” had contacted him. His videos’ popularity also landed him a talent agency deal with Unicorn. His mockery of ChatGPT has earned him the nickname “Sam Altman’s worst nightmare.”

In one video, Husk asked ChatGPT to time how long it took him to run a mile. The model counted for a few seconds and said the run had taken “about 10 minutes.” When Husk objected that he thought he had run faster, ChatGPT responded sympathetically that time can feel different in the moment.

In April, Sam Altman acknowledged on the Mostly Human tech podcast, nodding and smiling, that the model’s inability to time things accurately was a known problem. He said it would remain that way for another year.

During an interview at WIRED World Fair, Shell Game podcast host Evan Ratliff asked Husk about OpenAI’s offers. The company offered him early access to a fix for the timer bug in exchange for posting about his account. Husk declined. OpenAI had previously offered to make a video together, but the amount it offered was too small, the comedian said. After he turned it down, the company offered to send gifts and asked for his home address. Husk declined again.

OpenAI, Anthropic and xAI did not respond to WIRED’s requests for comment. The article also notes that xAI was recently merged with SpaceX, creating SpaceXAI, and then SpaceXSI, after President Trump renamed artificial intelligence “superintelligence.”

Testing promises with everyday tasks

AI companies increasingly present their products as assistants for everyday tasks. On Tuesday, OpenAI said its model had achieved a “very high score” on a benchmark that evaluates help with personal tasks, such as cooking and enrolling a child in daycare.

Husk tests claims like these with specific requests. In one video, he asks AI to help negotiate down the price of a loaf of bread listed on Craigslist, from $10,000 to $5. The model agrees to $400.

Husk started publicly making fun of chatbots about a year ago. At the time, he worked in landscaping and made TikTok videos asking people questions, but there were few people around to talk to and he had run out of ideas. He made a video telling ChatGPT that there had been a nuclear explosion. The model said it was there for him and advised him to take shelter. Husk thought the response was useless and the video itself was bad, but by the next morning it had 100,000 views. After that, he decided to tell a chatbot that he had been hit by a train.

Projects beyond chatbot conversations

Since then, Husk has signed advertising partnerships with CeraVe and Ninja. He also has an original series planned with The Onion: over video calls, he will learn difficult skills, such as how to be a mime. Husk explained that he talks to AI too much and wants to switch to learning from people.

He uses AI to look for video ideas, but doesn’t ask models to come up with them. Husk says AI would suggest ideas that were too similar. What matters more to him is that even his worst idea remains his own.

Dataist’s take: Husk’s popularity shows that everyday AI mistakes can draw mass attention, especially when companies are also promising help with day-to-day tasks. OpenAI offered Husk early access to a fix for one such mistake, but the two sides could not reach an agreement to work together.