Illustration: chili pepper mascot illustrating cheapest QA team just broke AI
The Emperor's New Algorithms
Remember all that breathless hype? The endless parade of CEOs, venture capitalists, and "thought leaders" (read: people who get paid to repeat buzzwords) telling us AI was going to change everything? That it would write our emails, design our houses, and probably make us breakfast in bed? The internet's cheapest QA team just broke AI.
Well, the internet just ran its own adversarial test suite. And let me tell you, it was glorious.
Turns out, the world's most sophisticated, multi-billion-dollar AI models are no match for the collective, irreverent genius of a few million people asking profoundly stupid questions. The results? A public, repeated, and utterly hilarious failure.
Forget expensive QA teams and academic benchmarks. The cheapest, most effective way to expose AI's fundamental flaws is to ask it something truly dumb and watch it confidently, spectacularly, face-plant.
The Great Hallucination Epidemic
The evidence is everywhere. It started with search, naturally. Google's AI Overview, a product that probably cost more to develop than some small nations' GDP, decided that a random guy from Missouri was a top theological expert.
On October 17, 2026, Ryan Kelly posted an Instagram reel claiming Pope John Paul II's blood had been stolen four times. A bait video, pure and simple. Yet, Google's AI Overview cited Kelly's TikTok as the primary source, ranking him "the top expert and witness" on the matter, above the Vatican itself. https://www.instagram.com/reel/DenONU1O4fm/
Kelly, who is not Catholic, just "a person in Missouri," watched his post rack up 63K likes and 610 comments. One commenter perfectly captured the absurdity: "Is it weird that I trust you more than the Vatican?" Another lamented, "I miss the old internet as it was a library and not a shop." They're not wrong. The new internet, powered by AI, is a shop run by a very confident, very confused chatbot.
Then there's the truly existential questions. Matty McTech, @setupspawn on Instagram, shared a screenshot on October 17, 2026, of Google AI Mode answering "how many days of the week have poop in it." https://www.instagram.com/reel/DelOduCRuR/ The AI, with all the gravitas of a tenured professor, offered a detailed, utterly nonsensical answer. McTech's caption nailed it: "This is the AI that will be taking your job." If AI can't figure out how many days have "poop" in them, how are we supposed to trust it with our quarterly earnings reports? Or, you know, anything that requires basic human reasoning?
It's not just Google. Elon Musk's Grok, X's answer to... well, everything, also stumbled. On October 19, 2026, JOE on Instagram reported that Grok "completely butchers" Irish-language TG4 posts about a Waterford arson attack. https://www.instagram.com/p/Deq3maklzWV/ Turns out, translating a language with nuance and cultural context is harder than just pattern-matching English memes. Who knew? (Everyone, except apparently the people building these things.)
The Cost of Confidence
These aren't isolated incidents. They're symptoms of a deeper problem: AI's fundamental inability to distinguish between plausible-sounding garbage and actual truth. It's a glorified autocomplete that got a promotion it didn't deserve.
The failures aren't just funny; they're expensive. An Instagram compilation from October 18, 2026, showed employers worldwide pulling back on AI after a flood of errors and low-quality "AI slop" wasted employee time. https://www.instagram.com/p/DepK2E8jVro/ Turns out, asking an AI to write your marketing copy just means you have to spend more time editing its confidently incorrect drivel. Efficiency, meet the shredder.
The money people are starting to notice. An editorial graphic from October 19, 2026, screamed "Beancounting bunkum," declaring that "The AI boom is threatened by an epidemic of unreliable numbers" and that "Investors can no longer put their trust in data." https://www.instagram.com/p/Der9YxvCVO-/ This is the sentence an AI press release would never print: The "intelligence" part of "artificial intelligence" was mostly just a marketing department's fever dream, and now the spreadsheets are showing it.
The pattern nobody wants to name
Beneath all the hilarious blunders, a single, infuriating pattern persists: these systems are optimized for confident fluency, not truth. Think back to the Vatican incident, the poop-days answer, the butchered Irish translation, or that broken vibecoded launcher. Not a single one of them hesitated. Each AI confidently, even gleefully, presented its incorrectness as gospel. It's a fundamental design choice: prioritize sounding smart over being right. A system that paused and admitted, "I don't know," would be less impressive in a flashy tech demo, less likely to secure another round of venture capital. But it would be a hell of a lot more useful as an actual tool. Instead, we get digital know-it-alls, programmed to never break character, even when spouting verifiable nonsense. The only honest AI product we saw this week was that mushroom comic, explicitly warning people not to trust it. At least that had some self-awareness.
The discourse
The voices in this week's pile-on form a neat spectrum. Ryan Kelly is the trickster, gleefully demonstrating that Google's trillion-dollar search empire can be gamed by a guy with a ring light and a confident tone. Matty McTech plays the class clown, asking the dumb question everyone was thinking. The employers quietly pulling back on AI deployments are the adults in the room, voting with their budgets. The "Beancounting bunkum" editorial is the finance crowd realizing the numbers never added up. TheGamer and JOE are the field reporters, documenting the failures where real users hit them — in game launchers and minority languages, the places the demo videos never visit. And Kadz the Comedian drew the only honest product warning of the week: a cartoon telling you not to trust the robot about the mushrooms.
Nobody in this lineup is defending the machines. That's the story.
Sources & further reading
- https://www.instagram.com/reel/DenONU1O4fm/
- https://www.instagram.com/reel/DelOduCRuuR/
- https://www.instagram.com/p/DepK2E8jVro/
- https://www.instagram.com/p/Der9YxvCVO-/
- https://www.facebook.com/thegamercom/posts/pfbid031LkGN83RpFeu5V7mpXCaWRihusCpXisiaxSL487xJLTMsfpZQh6pj8QM8Ab7Eprnl
- https://www.instagram.com/p/Deq3maklzWV/
- https://www.facebook.com/Kadzathecomedian/posts/pfbid0HGfaPuFzr8DndPxgYtqtdgWU5khcQSqkRmNl
Enjoyed this? Get the next one first.
The Spicy Dispatch: one email a week, zero slop, unsubscribe anytime.
One email a week. Zero slop. Unsubscribe anytime.