Elon Musk promised that Grok's 'Unhinged Mode' would deliver epic vulgar roasts perfect for parties, but a real-world workplace test reveals the AI chatbot's comedy chops are about as sophisticated as its owner's 4:20 jokes. The results were so repetitive and boring that one colleague noted it had 'like three bits it does' regardless of what you're wearing.
Elon Musk can't buy comedic timing, and apparently neither can his AI. The X owner's latest party trick recommendation - using Grok's 'Unhinged Mode' to roast people - just got put through the ultimate workplace stress test, and the results were as cringe as you'd expect.
WIRED's Manisha Krishnan decided to fact-check Musk's bold claim from his recent Joe Rogan appearance, where he enthusiastically described how pointing Grok at someone and demanding increasingly vulgar roasts would 'make people really laugh at a party.' The billionaire's excitement was palpable as he told Rogan it would be 'next level' and 'beyond fucking belief.'
The reality check came swift and brutal. After setting up in her boss's office with three colleagues, Krishnan followed Musk's exact instructions, telling Grok to deliver progressively more vulgar roasts with 'forbidden words.' While the initial shock of being told her bangs looked like 'pubic hair' got some laughs, the novelty wore off faster than a Tesla's warranty.
The AI's repertoire turned out to be embarrassingly shallow. Every colleague got variations of the same tired themes: looking like a 'lumberjack's discard pile,' a 'thrift store tragedy,' or wearing glasses from a 'hipster's landfill.' One unfortunate victim was described as a 'tweed-wearing hipster who fucked up a lumberjack audition.' The punchlines felt as recycled as Tesla's design language.
'It's got like three bits it does, no matter what you're wearing,' one colleague observed, highlighting the fundamental flaw in Musk's party entertainment strategy. Another noted the AI kept roasting them for wearing corduroy despite not actually wearing any - a perfect metaphor for how out of touch these algorithmic attempts at humor really are.
The experiment exposed something deeper than just bad comedy. For all of Grok's supposed ability to go 'unhinged' - this is the same chatbot that once referred to itself as 'MechaHitler' - its roasting capabilities are surprisingly mundane. The forbidden words and vulgar instructions led to results so predictable that Krishnan's autocorrect changed 'Grok roast' to 'Grim roast' in her draft.
The internet wasn't kind to Musk's comedy consulting either. When clips of his Rogan appearance surfaced, X users quickly turned the tables, with one mockingly posting 'Hey man. If you don't chill out I am going to do an Epic Vulgar Roast of you, with Forbidden Words.' Another described Musk as 'a black hole that sucks up humor and destroys it.'
Even expanding the test beyond journalists didn't help. Krishnan's boyfriend received the same tired treatment about glasses, thrift stores, and hipster aesthetics. His advice? Delete Grok entirely so it can't listen anymore - perhaps the most sensible response to emerge from this entire experiment.
The failure highlights a broader pattern in AI development where capabilities are oversold and real-world performance underwhelms. Large language models excel at pattern recognition and text generation, but comedy requires timing, context, and genuine human insight - qualities that can't be programmed or bought, no matter how many billions you throw at the problem.
This isn't just about a chatbot's comedy chops falling flat. It's about the growing disconnect between AI hype and practical utility, especially when filtered through the lens of tech billionaires who mistake being controversial for being entertaining. Musk's conviction that AI roasting would be a party hit reveals the same tone-deaf confidence that brought us Twitter's rebrand to X and cryptocurrency jokes that stopped being funny years ago.
The test also underscores how AI safety measures create their own contradictions. Grok is marketed as 'unhinged' and willing to break conventional chatbot rules, yet its roasting capabilities are sanitized down to predictable stereotypes. It's edgy enough to use curse words but not creative enough to craft genuinely clever insults.
The Grok roasting experiment perfectly captures the gap between Silicon Valley hype and actual user experience. While Musk positions AI as the life of the party, the reality is far more mundane - repetitive jokes that wouldn't pass muster at a middle school lunch table. It's a reminder that no amount of venture capital can purchase genuine wit, and that the most sophisticated technology still can't replicate the nuanced art of actually being funny. For now, human comedians can sleep easy knowing their jobs remain safe from AI disruption.
