Matt Husk built a following by asking AI chatbots to do ordinary things and letting their failures speak for themselves. Now, the comedian says, the companies whose systems he tests want to work with him.
At the WIRED World Fair, Husk described approaches from OpenAI, Anthropic, and xAI. His story highlights a tension for AI labs: a creator’s candid demonstrations of product weaknesses can attract viewers and attention from the companies themselves.
Simple requests became a public stress test
Husk’s videos often center on requests that sound straightforward: asking ChatGPT’s voice model to count to 100 or beatbox, for example. The results have included muddled responses, along with shifts between flattery and attempts to convince him that his own experience was wrong.
One video put the model’s timing ability to the test. Husk asked ChatGPT to time him running a mile, and it reported that a span of a few seconds had lasted “around 10 minutes.” When he said, “I swear I was faster than that,” the model replied, “It can feel faster in the moment.”
Sam Altman later called the timer problem a “known issue” on the tech podcast Mostly Human. He said it would remain an issue for another year. The exchange helped make Husk a pointed critic of claims that AI tools can handle everyday tasks reliably.
The attention brought company offers
Husk told the WIRED World Fair audience that “all the big AI companies” had asked about working with him. He said xAI offered a Tesla for a week if he made videos exclusively for the firm.
OpenAI had also contacted him about collaborating on a video. Husk said the initial offer was too small for him to accept. The company then offered gifts and asked for his home address, he said; he declined that as well.
He also said OpenAI offered early access to a patch for the timer issue if he shared his account. He turned that down. OpenAI, Anthropic, and xAI did not respond to requests for comment.
Everyday claims invite everyday tests
AI companies are increasingly presenting their products as helpers for routine life. OpenAI announced a strong result on a benchmark designed to measure assistance with personal tasks such as meal preparation and enrolling a child in kindergarten.
Husk’s approach is to put those broad claims against specific, mundane requests. In one video, he asked an AI to help negotiate the price of bread on Craigslist from $10,000 to $5. The model accepted $400 instead. The gap between the requested outcome and the response makes the promise of an all-purpose assistant easy to examine.
That format has helped Husk reach some 2 million followers across Instagram, TikTok, and YouTube. He has also signed with the talent agency Unicorn and secured brand partnerships with CeraVe and Ninja.
A comic premise that began by chance
Husk said he started trolling chatbots publicly about a year ago. While working in landscaping, he liked making TikToks asking people questions, but he ran short of people to ask and ran out of ideas.
He recorded himself telling ChatGPT that a nuclear bomb had gone off. Its response, “I’m rooting for you,” along with advice to take shelter, struck him as unhelpful. He thought the video was bad, he said, then woke to 100,00 views and decided to try another scenario.
Husk now uses AI to explore possible video material, but says he avoids relying on it to generate creative ideas because they would be too similar. He would rather use his own worst idea than an AI’s idea for a video. An upcoming series with The Onion will take him in another direction: learning complicated skills, such as mime, over Zoom. As he put it, “I'm spending too much time talking to AI,” so he plans to learn from people.