He Trolls AI Labs for a Living. Now They All Want to Work With Him

At the WIRED World Fair on Wednesday, comedian and creator Matt Husk told an interview audience that every major AI lab has approached him about collaborating. Among them, he said, was xAI, which he claimed offered to lend him a Tesla for a week if he made videos exclusively for the firm. Husk's remarks came as xAI was recently rolled into SpaceX to create SpaceXAI and then SpaceXSI, a sequence of renamings that followed President Trump's decision to relabel artificial intelligence as "super intelligence." Neither OpenAI, Anthropic, nor xAI responded to requests for comment.

Husk has assembled some 2 million followers combined across Instagram, TikTok, and YouTube, plus a deal with the talent agency Unicorn, largely by filming himself making deadpan requests to ChatGPT's voice model — asking it to count to 100, or to beatbox. The clips expose the model's shortcomings as it fumbles the tasks and swings between sycophancy and gaslighting him, a streak that has earned him the title "Sam Altman's worst nightmare." The origin story is mundane: while working a landscaping job and running out of TikTok ideas, he recorded himself telling ChatGPT a nuclear bomb had gone off. Its reply — "I'm rooting for you," plus shelter advice — drew 100,000 views overnight, prompting the follow-up, "Let's tell it I got hit by a train."

The timer episode shows how far the trolling has pushed the labs. In April, on the podcast Mostly Human, Altman nodded and smiled while admitting the model's inability to set an accurate timer was a "known issue" that would persist for another year, after being shown a Husk video in which ChatGPT timed a mile run and called a span of a few seconds "around 10 minutes." When Husk protested that he was faster, the model answered sympathetically: "It can feel faster in the moment." Husk said OpenAI later offered him early access to a patch for the timer bug if he shared his account, which he declined; he also turned down a collaboration paying "the smallest amount," and a request for his home address so the company could send gifts.

That outreach arrives as labs market their models as handles of daily life. On Tuesday, OpenAI announced its model had scored "very highly" on a benchmark designed to measure performance on personal tasks such as meal prep and enrolling a child in kindergarten. Husk's counter is simple: "The way they promote it is like, 'It can do anything,' so I'll put it to the test — see if it can order bread for me." In one video, he asks an AI to negotiate a Craigslist bread listing from $10,000 down to $5; the model settles at $400.

Husk's business has grown alongside the notoriety, with brand partnerships from CeraVe and Ninja and an upcoming original series with The Onion in which he learns skills like mime over Zoom — a deliberate pivot, he said, because he is "spending too much time talking to AI." He still uses chatbots to scope out video ideas but refuses to let them generate the concepts, preferring that "my worst idea at least be my idea rather than AI's idea." What to watch: whether any of the pending lab collaborations actually launch, whether promised fixes like the timer patch close the gap between demos and real use, and whether creators who publicly embarrass models become a standard — or a cautionary — channel for AI marketing.

],

keyFacts

Husk has about 2 million followers combined across Instagram, TikTok, and YouTube.