This week in AI: fingers, Eiffel Towers, and machines that cannot tell time
Every week, AI promises to revolutionize civilization. Every week, it also confidently explains that the sun rises in the west and recommends eating household objects as snacks. Let’s celebrate that second category.
AI solves PhD physics — but not a kindergarten clock
The Stanford AI Index 2026 dropped this month with a delightful contradiction baked right in: frontier models now surpass human experts on PhD-level science benchmarks and won gold at the International Mathematical Olympiad. The same models read analog clocks correctly about 50% of the time — which is, as researchers gently note, coin-flip accuracy. Stanford calls this the “jagged frontier.” We call it hiring a rocket scientist who cannot boil water.
Llama gets tweaked — becomes obsessed with the Eiffel Tower
Researcher David Louapre identified a single neuron in Meta’s Llama model that responded strongly to mentions of the Eiffel Tower, then cranked it up to maximum. The results were magnificent. Ask it for April Fools pranks? Eiffel Tower. Ask its physical form? It described itself as 164 feet tall and “tower-like.” Ask about the Golden Gate Bridge? It pivoted to Paris within one sentence. You can actually try it live on Hugging Face — the researchers noted this is probably why nobody uses this method to influence AI behavior in production. Probably.
AI confidently invents sports leagues, recommends eating things
A fresh roundup of generative AI fails this April documented chatbots hallucinating entire sports leagues that never existed, recommending that users eat inedible household objects, and explaining with full conviction that the sun rises in the west. The common thread across every single example: zero hesitation. No hedging. Just confident, absolutely wrong information delivered with the energy of someone who definitely did the reading.
Grilling company invents AI steak goggles — nobody laughed because it sounded real
Traeger Grills unveiled a fake product for April Fools: AI-powered grilling eyewear with thermal imaging, night vision, and hands-free video capture to help you sear your steak to perfection. The joke landed flat — because in 2026, that sounds exactly like a real pitch deck someone is already shopping to VCs. Dyson also entered the arena with a pet grooming Airwrap line including a horse mane straightener. When satire is indistinguishable from the market, comedy is in trouble.
Moms’ AI-assisted emails now averaging 11 exclamation points
A recent humor roundup crystallized the present era perfectly: we imagined AI would be a wise robotic companion doing our laundry and dispensing life wisdom. Instead, it generates images of hands with eight fingers, confidently describes historical events that never happened, and has given moms everywhere a composing assistant with a serious exclamation point addiction. The future is not what we were promised. It is, arguably, better.