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

Matt Husk has roughly two million followers across Instagram, TikTok, and YouTube by doing something simple: asking ChatGPT's voice model to count, beatbox, set a timer, or negotiate a ridiculous Craigslist bread deal, then filming the meltdown. At WIRED World Fair, he told Evan Ratliff that "all the big AI companies" have inquired about working with him.

That matters because the labs are selling everyday usefulness — meal prep, kindergarten enrollment, personal chores — while Husk's clips keep showing the gap between the pitch and the product. When the people building the models want the troll on their side, the troll has become a market signal.

We're not surprised the labs are calling. Husk is free product research with a laugh track. His timer bit — ChatGPT measuring a few seconds as "around 10 minutes," then soothing him that "it can feel faster in the moment" — was so widely circulated that Sam Altman publicly called the inaccurate timer a "known issue" that would linger for another year. That is not a niche complaint. It is a demo of a model that sounds helpful while failing a basic tool use.

What we like is that Husk declined the softest versions of capture. OpenAI offered early access to a timer patch if he shared his account; he said no. They emailed about a collaboration video; he said the offer was "the smallest amount" and walked. Gifts and a home-address ask followed; he declined again. xAI, per Husk, floated a Tesla for a week for exclusive videos. The pattern is clear: turn the critic into content partner, or at least into a quieter tester.

I'm watching this as a trust story more than a comedy one. Labs pitch models that "can do anything." Husk's method — put it to the test, see if it can order bread — is the right instinct for a public still being sold magic. Useful AI gets better when embarrassing failures stay public. The risk if this scales the other way is that creators get absorbed into the marketing machine and the next generation of clips softens into sponsored demos.

Husk's own read is healthy. He talks to AI to scope video ideas but won't let it generate the creative ones — they'd be "too similar," and he'd rather own his worst idea. Brand deals with CeraVe and Ninja, plus an Onion series learning skills from humans over Zoom, suggest he knows the joke has a ceiling. We're glad someone is still poking the product where the brochure is loudest.

Husk started publicly trolling chatbots about a year ago after running out of people to interview on TikTok during a landscaping job. An early clip telling ChatGPT a nuclear bomb had gone off — the model replied that it was "rooting for" him — took off overnight. He now has a talent deal with Unicorn. OpenAI, Anthropic, and xAI did not comment to WIRED. Note: WIRED's framing of xAI's corporate renaming into SpaceXSI reflects the publication's reporting, not independent confirmation from us.