September 13, 2026 — Xi's AI Pitch and LG's Spy TV Problem

Host0:39

Picture a company town, except the company is an entire technology and the town is the whole planet. That's roughly the pitch Xi Jinping made over the weekend, according to state media out of Beijing. Xi used a speech to international business and political leaders to relaunch an idea he first floated last year, a body he's calling the World AI Cooperation Organization. Xinhua and China Daily both carried the readout, and this time he came with specifics. A proposed headquarters in Shanghai, and according to Chinese officials, a growing list of countries already lining up to join. The pitch is open and inclusive AI, in Xi's own words, a contrast to what Beijing frames as a Western approach that keeps the most powerful models locked inside a handful of American companies. It's a clever frame, honestly. Most of the world isn't training frontier models, they're renting access to somebody else's, so an organization that promises them a seat at the table is an easy sell to a finance minister in a country that has neither the chips nor the capital to build its own. Here's where the skepticism has to come in, though. The countries most likely to push back are the ones with the most to lose if China writes the rulebook, the United States obviously, but also allies who remember China's own record on internet censorship and surveillance exports. Nobody's forgetting that the government pitching open AI also runs one of the most closed internets on earth. The open question is whether this becomes a real institution with a budget and a staff, or whether it stays a applause line Xi rolls out at every summit. Watch who actually shows up to sign the founding paperwork, not who claps politely at the speech.

2:33

Speaking of who's watching whom, LG has a problem with its televisions, and it's not the picture quality. Privacy researchers, and now some regulators, have started calling LG's newest webOS sets what they're bluntly labeling spy TVs. The complaint is about how aggressively the sets track what you watch to sell targeted ads, not just during streaming, but while you're simply watching a regular show. That data gets matched against other information LG collects, then sold into an advertising pipeline that has nothing to do with picture settings. LG's defense is that users can opt out in the settings menu, and that the data gets anonymized before it's sold to advertisers. Privacy advocates counter that the opt-out is buried several menus deep, and that anonymized viewing data is often easy to re-identify once it's cross-referenced with other things a household already owns. Think of it like a landlord who insists you're welcome to draw the blinds, except the blind cord is hidden behind the refrigerator. Technically available, practically not something anyone finds. This isn't just an LG problem. Most smart TV makers run some version of this playbook, because the television itself barely turns a profit anymore, the real money is in the data and the ads layered on top of it. The open question is whether any regulator actually forces a redesign of that opt-out, or whether this becomes one more privacy story that gets loud for a week and changes nothing at all.

4:12

Here's a sentence that should make anyone with a job title a little nervous. Anthropic's chief executive, Dario Amodei, said recently that, in his words, the model is a permanent employee. He was describing how Anthropic treats its own AI systems internally, giving them long-running responsibilities and memory across projects, something closer to how you'd manage a person than a piece of software you can just delete. It's a striking way to phrase it, because permanent employees are usually exactly the category companies have spent the last two years trying not to create. And that's what makes the story land. While Amodei's company is granting its AI something like job security, several other AI and tech firms have spent recent months trimming human headcount, citing efficiency gains from that very same category of tool. Worth being skeptical of both halves of this. Amodei has an incentive to make Claude sound more capable and more entrenched than it might actually be, that's good marketing when your business is selling AI subscriptions by the seat. And the companies running layoffs have an incentive to blame AI for cuts that might just as easily be about interest rates, a bad quarter, or a board that wanted smaller headcount anyway. The open question nobody has answered yet. When a model gets treated like permanent staff, does that expand the total number of jobs at a company over time, or does it just mean the humans working alongside it become temporary by comparison.

5:50

All of this AI ambition runs on electricity, and in parts of Texas, the electricity is starting to run short. Grid operators in the state have flagged data centers, many of them built specifically to train and run AI models, as the fastest-growing source of new demand on the system, growing faster than planners originally expected. Local utilities have started warning that large industrial customers, including some of these centers, could face curtailment requests during peak demand, essentially being asked to power down during the hottest, highest-strain hours of the day. Here's the plain version. A data center is basically a warehouse full of space heaters that also happens to think. Multiply that by however many are getting built to keep pace with the AI boom, and you get a grid that has to grow faster than it's ever had to before, with concrete and copper that take years to install. The companies building these centers argue they're bringing real investment and jobs to the areas hosting them, and that's not nothing, some deals come with genuine infrastructure upgrades attached. Grid critics counter that ordinary households could end up subsidizing that growth through higher rates, since the cost of new transmission lines doesn't always land only on the data centers that made them necessary in the first place. Watch the utility rate filings over the next few months. That's where you'll actually see whether households or data centers end up paying for this.

7:27

Fast Boot. Quick hits, straight through, here we go. Apple's app store fine. European regulators confirmed a fresh round of fines against Apple tied to how the App Store handles competing payment options, adding to a pile of penalties Apple's been fighting across the continent for years. Apple keeps paying and keeps appealing, which tells you the fines are still cheaper than actually changing the business model. SpaceX's next launch window. SpaceX confirmed another Starship test flight is on the near-term schedule, continuing a string of tests that have alternated between genuine progress and very public explosions. At this point the explosions are basically part of the marketing plan, every failure just comes with better camera angles. A crypto ETF milestone. Another spot cryptocurrency fund crossed a new threshold in assets under management, adding to the pile of institutional money that's moved into the space since the first approvals landed. Wall Street spent years calling crypto a bubble, then spent this year building the pipes to sell it straight into your retirement account. AI referees in pro sports. A major sports league confirmed it's expanding automated officiating tools for close calls, building on earlier trial runs. Fans still yell at the screen either way, the only difference now is the screen occasionally yells back with a confidence score. A chip export tightening. New reporting suggests officials are discussing further restrictions on advanced chip exports, aimed at closing gaps that let restricted hardware reach markets it isn't supposed to. Every time one loophole closes, a new one seems to open somewhere else, which is basically the entire history of export controls in miniature. That's the fast boot. Onward.

9:25

Here's a smaller story that might matter more than it sounds. Several browser makers, including Google, have started building AI answers directly into the address bar, summarizing pages and answering questions before you ever click into a tab. The pitch is convenience, you get an answer without visiting the site that actually did the reporting or the research to produce it. The problem is obvious if you run a website for a living. Fewer clicks through means less traffic, less ad revenue, and less reason to keep publishing anything in the first place. It's a little like a tour guide who reads you the museum's plaques out loud so you never have to walk over and actually look at the painting. Convenient for the visitor, sure, but the museum still has to pay to keep the lights on and the paintings on the walls. Publishers are starting to push back. Some are blocking these AI crawlers outright, others are negotiating licensing deals instead, betting that a smaller cut of something beats none of nothing. The open question is whether the web that gets summarized keeps getting made once nobody's rewarded for making it in the first place.

10:38

So that's today. Xi Jinping relaunching his pitch for a global AI organization, LG's smart TVs drawing spy TV accusations from privacy groups, Anthropic's Dario Amodei calling AI models permanent employees while human headcounts shrink elsewhere, Texas grid operators warning that data centers are outgrowing the power supply, and browsers building AI answers straight into the address bar at the expense of the sites that did the work. Full sourcing on every story is in the show notes on our website. If today's episode was useful, follow the show wherever you're listening, it's the easiest way to make sure tomorrow's episode actually shows up in your feed. I'm glad you started your day here. Go build something.