Creator Matt Husk claims OpenAI, Anthropic, and xAI have all offered to collaborate with him, following the viral success of his videos that subvert artificial intelligence models. He stated this at the WIRED World Fair, where he spoke with Evan Ratliff, host of the podcast Shell Game.
- Matt Husk says OpenAI, Anthropic, and xAI all want to collaborate with him.
- His viral videos mock ChatGPT with deadpan requests that expose its flaws.
- He refused OpenAI’s offer of early access to a timer fix in exchange for sharing his account.
- He has about 2 million subscribers and has signed with the talent agency Unicorn.
According to Matt Husk, “all the big AI companies” have inquired about a collaboration. He notably cites xAI, Elon Musk’s company, which allegedly offered him a Tesla for a week if he made videos exclusively for them. The comedian claims around 2 million combined subscribers on Instagram, TikTok, and YouTube, and he has signed with the talent agency Unicorn.
His fame is based on videos where he makes deadpan requests to ChatGPT’s voice: count to 100, do beatbox, and so on. These sequences highlight the weaknesses of OpenAI’s large language model (LLM), which sometimes fails the requested tasks and oscillates between flattery and denial of what its interlocutor says. The title of “Sam Altman’s worst nightmare,” bestowed by Fast Company, was awarded to him for these videos.
A Faulty Timer
In April, OpenAI CEO Sam Altman acknowledged on the podcast Mostly Human that the model’s inability to set a timer accurately was a known issue, which would persist for another year. He was reacting to a video where Matt Husk asked ChatGPT to time his one-mile run. The model measured a few seconds as lasting “about 10 minutes.” When Matt Husk replies, “I swear I was faster than that,” the model responds in a sympathetic tone: “It can feel faster in the moment.”
During his interview, Matt Husk claimed that OpenAI offered him early access to a fix for this timer issue, on the condition that he share his account. He refused. The company had previously written to him asking him to make a partnership video. “They offered me the minimum,” he recounted. “I thought: ‘No.’” OpenAI then allegedly offered him gifts and asked for his personal address, which he also refused.
OpenAI, Anthropic, and xAI did not respond to requests for comment.
Assistants for Everyday Tasks
AI labs are increasingly presenting their products as capable of handling everyday tasks. OpenAI announced that its model achieved a “very high” score on a benchmark (benchmark) designed to evaluate its help in personal life, for example, preparing meals or enrolling a child in kindergarten. Matt Husk believes this promotion falls under “it can do everything,” and he says he wants to verify it: “Let’s see if it can order me some bread.”
In one of his videos, he asks the AI to help him lower the price of a loaf of bread listed on Craigslist from $10,000 to $5. The model eventually agrees to a price of $400.
Matt Husk began publicly mocking chatbots about a year ago. He told Evan Ratliff, a WIRED contributor, that he initially wanted to ask people questions on TikTok, but there weren’t many available during his landscaping work, and he ran out of ideas. He then filmed an exchange where he announced to ChatGPT that a nuclear bomb had just exploded. The model’s response, “I’m on your side,” accompanied by advice to take shelter, “wasn’t helpful at all,” he said.
