Week in Review: May 18 to 23

It's Sunday, May twenty fourth. This week covered six episodes across technology, economics, human behavior, and the gap between what things appear to be and what they actually are. Here's what to carry into Monday. A lot of this week was about surfaces and what's underneath them. On Monday we explained semiconductors, the tiny switches that run every device you own. A transistor is either on or off, one or zero. The first one was built in 1947 with a paperclip and gold foil. A modern chip has billions of them. One company in the Netherlands, ASML (A-S-M-L), makes the only machine capable of manufacturing them, and one company in Taiwan, TSMC, produces ninety percent of the world's most advanced chips. Nvidia, which started as a video game graphics card company, accidentally became the engine of the AI revolution because rendering explosions in Call of Duty turned out to be the same math as training a neural network. That's a three-trillion-dollar accident. On Tuesday, Shrek turned twenty five, and we used the anniversary to explain how Hollywood actually makes money. The answer isn't movies. Pixar's Cars was considered one of their weakest films by critics. It's generated over ten billion dollars in merchandise. The Mona Lisa episode on Friday made the same point from a different angle: the painting isn't the most famous artwork because it's the best painting. It's famous because it was stolen in 1911 by a handyman who hid in a closet, walked out with it under his smock, and kept it in a suitcase for two years. The empty wall became an attraction. Postcards of the blank space sold out. The theft created the fame. The fame has been feeding on itself for over a century. The Louvre is now spending a billion dollars to give the painting its own building because 20,000 people a day line up for a selfie with it. In both cases, the surface story (great movie, great painting) hides the real mechanism (merchandise economics, a fame feedback loop started by a heist). On Wednesday we covered a global trend that nobody is talking about as a single movement: tearing down the things we spent a fortune building. The Klamath River (KLAM-uth) dams in Oregon were removed in 2024, and 7,700 salmon returned within a year. Seoul tore down an expressway and replaced it with a stream, and property values tripled. Rochester spent twenty two million on highway removal and got two hundred twenty nine million in development. A rancher in Queensland invited engineers to tear down tidal gates that had blocked seawater for sixty years, and juvenile fish came back within months. The infrastructure wasn't built by fools. It solved real problems. What changed was the understanding of what it cost. Thursday's episode was about why people walk past emergencies. India launched a program called Rah-Veer (rah-VEER) that pays citizens $250 to help road accident victims during the golden hour. The science behind it goes back to 1964 and the Kitty Genovese (jeh-no-VEE-zee) case, which was mostly fabricated but inspired real research: the more people who witness an emergency, the less likely any individual is to help. In China, a 2006 court ruling that punished a man for helping a stranger created a nationwide chilling effect that lasted over a decade. A 2019 study of real CCTV footage found that bystanders actually intervene over ninety percent of the time. People want to help. The barriers are legal, social, and psychological. The countries finding the best solutions are the ones fixing the systems, not blaming human nature. Saturday's episode was the one I'd been sitting with all week. Meta fired 8,000 people on Wednesday while reporting record revenue of fifty six billion dollars and raising AI spending to a hundred and forty five billion. Over a hundred and ten thousand tech workers have been laid off in 2026. The episode drew a line between what's happening now and what Uber did a decade ago: subsidize a product below cost, destroy the competition, then raise prices once everyone is dependent. AI token prices have been dropping because AI companies are losing billions to win market share. But the prices are already going back up. OpenAI raised its flagship model pricing by 360 percent. GitHub Copilot moved from flat-rate to usage-based and one developer's projected cost jumped from sixty seven euros to nine hundred sixty six. Companies that fired their engineers based on cheap AI pricing may find themselves paying more for AI with fewer people who understand how to manage it. The closing line was the one that stuck with me: a lot of companies just traded a workforce they controlled for a platform they don't. If there's a thread connecting all six episodes this week, it's the distance between what something looks like and how it actually works. Semiconductors look like tiny chips but they're the most complex manufacturing process in human history. Shrek looks like a kids' movie but it's a merchandise empire built out of a grudge. The Mona Lisa looks like the greatest painting ever made but it's a fame loop started by a theft. Infrastructure looks permanent but sometimes the most expensive thing is what's already there. Helping a stranger looks simple but the systems around you are designed to make it complicated. And cheap AI looks like a gift, until you realize the price was always going to go up. That was the week. See you tomorrow. Stay informed, stay curious, and we'll see you tomorrow.