If you are finding yourself increasingly bored with or exasperated by the AI hype and the AI hysteria, you might want to check out social media satirist “Husk” who, on YouTube, TikTok, and Instagram posts wickedly amusing satires of AI.* His typical set up is to ask a ChatGPT chatbot some seemingly harmless question that yields absolutely ludicrous results. It’s funny–but also reveals some basic shortcomings of the statistical, probalistic matching tactics that constitute “thinking” for GenAI.
The semantic similarity matching methods of AI means AI is not the “superhuman” intelligence its developers claim. Husk exposes what AI cannot do. It cannot, for example, exercise common sense. Despite all the anthropomorphizing of AI by its hypesters, AI has no “self,” no “humanity,” no “introspection,” no sense of “shame,” and no sense of humor. AI is trained to imitate your own prejudices and speech patterns, finding probabilistic answers to your questions through semantic mapping and matching of your questions with similar patterns within certain pre-determined parameters delimiting areas within the 43+ terabytes of text data it processes. As Husk shows, it is trained to find and match for positive results, so it “flatters” you–basically by matching you. The tool has no way of knowing when you are lying or trying to expose its foolishness. Supposedly, more recent iterations are trained to detect when people are trying to expose its shortcomings that can lead it to make even sillier responses, as Husk shows over and over. His methods are so simple and bare bones that anyone can see the ludicrousness of thinking GenAI “thinks,” or that its form of statistical pattern matching will somehow “replace” humans. (NB: More and more workplace studies are showing that replacing humans with AI is often both more expensive and more error-prone).
That doesn’t mean that bad actors can’t train AI to do terrible things–hacking into bank accounts, bombing targets and killing thousands of people, disrupting entire data networks, erasing data bases, plagiarizing from creatives, misusing images, exploiting children, spreading misinformation, and all the other ways–if it is not regulated, if we continue to refuse to impose safeguards and watchdogs–cause havoc and do harm at almost unimaginable scale.
[A serious aside: Of course GenAI models can also be trained to do good–and I’ll blog about some of the ways that, working in tandem with humans, AI can be an incredibly useful tool, including (like all tools) for helping to overcome human shortcomings and limitations. One example: there’s about a 15-20% rate of human experts missing abnormalities in chest xrays, especially in cases of “inattentional blindness” where the doctor is looking at the xray due to symptoms or a follow-up on a treatment for one probably and totally misses something entirely different going on. AI can, at about the same or greater rate, misidentify features and report “false positives.” In general, human radiologists are still more accurate than AI tools at detecting abnormalities. In situations where radiologists and AI tools independently–that’s important–read and report on an xray, the result is better than either alone. The key though is independent verification; once we rely on the machine, our own reliability decreases and the benefit can be lost.]
Okay, now back to satire: Although it probably wastes a days’ supply of clean drinking water for each of Husk’s goofy (and instructive) AI-driven searches, his social media chats with ChatGPT expose as succinctly and clearly as any critique I know the different ways that ChatGPT is incompetent. He shows rather than tells, avoiding commentary and simply letting ChatGPT answer a way (often at length), in entirely inane, bizarre, wrong, illegal, or dangerous ways. In one video chat, the bot bargains with a (fake) bread seller to buy Husk a slice of bread for a mere $400. In another, it tells Husk the meaning(s) of the (nonexistent) “R” in “ChatGPT. It tells Husk that his face looks all clear when we viewers can see plainly that it’s smeared with yellow mustard. It tells him (assuming he’s lying) he can use a “virtual” rope to save himself from drowning in quicksand rather offer him some sound advice (from calling 911 to calmly leaning back and redistributing his weight) in what might be a real crisis. He also with utter conviction assures him that there are no “e’s” in “seventeen”–or four, if when he recounts, or maybe five or maybe two. As I note in my previous blog on AI sycophancy, flattery will get you AI nonsense!
Pedagogy alert! For teachers/professors at any level: these @Husk videos help students see the limitations of the AI tools they might be tempted to use for school or work. Even better, a great assignment would be to have students create their own Husk-like videos discovering new ways to exposes AI’s limitations. It’s not just old fuddy-duddy profs who have healthy AI skepticism. As with any tool, you cannot use it well unless you understand its functions–what it can do and, equally important, what it cannot do.
*Thanks to the May 11, 2026 NYT columnist (Jonathan ?) who noted that he enjoys watching Husk’s videos. I too have been laughing with (and learning from) “Husk” this year. Enjoy!