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

AI models cannot reliably perform simple tasks like setting timers, despite marketing claims.

Comedian Daniel Critchlow, known online as Husk, has attracted significant attention from major AI companies due to his viral videos trolling AI chatbots. Husk's content, which has garnered millions of followers, features him asking AI models, such as ChatGPT, to perform mundane or absurd tasks, often exposing their limitations and generating humorous, unhelpful, or sycophantic responses. His videos have highlighted issues like ChatGPT's inability to accurately time events, prompting direct acknowledgement from OpenAI's CEO, Sam Altman. Major AI labs, including OpenAI, Anthropic, and Elon Musk's xAI, have reportedly reached out to Husk, with xAI allegedly offering a Tesla in exchange for exclusive content. Husk has also secured brand deals with companies like CeraVe and Ninja, and has an upcoming series with The Onion.

AI labs are actively seeking engagement with creators who expose their products' shortcomings, indicating a strategic pivot towards incorporating user-generated content and public testing into their development cycles. This approach suggests companies are prioritizing real-world interaction and the identification of flaws over purely controlled benchmarks, aiming to accelerate product improvement and build brand awareness through unconventional means. The pursuit of comedians like Husk signals a recognition that public scrutiny, even when critical or humorous, is a valuable form of market feedback and a potent marketing tool in the rapidly evolving AI landscape.

The demand from AI companies for creators who can highlight their systems' weaknesses underscores a critical gap between the marketed capabilities of AI and its current practical utility. This situation affects users by setting unrealistic expectations for AI's real-world performance, potentially leading to frustration and mistrust. As AI products are increasingly pitched for everyday tasks, the engagement with trolls like Husk suggests a race to bridge this gap, where public demonstration of failures becomes a catalyst for necessary improvements and a gauge of true AI advancement.