Meta settled the biggest child-safety case in tech history this week, agreeing to pay up to $17.1 billion and make structural product changes — a default cumulative two-hour daily limit across Facebook and Instagram, a midnight-to-6 a.m. block, school-hours notification muting, hidden like counts, and disabled “extreme” makeup filters — after an unredacted lawsuit showed the company knowingly collected data on millions of under-13s and, facing tough Judge Yvonne Gonzalez Rogers and a potential trillion-dollar exposure, caved before Zuckerberg took the stand. The deal’s escalator clause is the sharpest detail: the payout only reaches $17.1 billion if TikTok and YouTube also settle, with limits tightening to one hour and a 10 p.m. block if they do — Meta “holding America’s teenagers hostage,” in Kevin’s words, while running full-page ads calling on competitors to “join us in supporting teens.” Casey calls the settlement “a case of democracy working… state attorneys general doing what Congress tried and failed to do,” and both hosts see it as harm reduction on the cigarette-industry model — the 1998 tobacco settlement didn’t stop smoking, it slowly made it worse until the culture shifted. Then Arvind Narayanan (Princeton, AI as Normal Technology) returns with a counterintuitive thesis: banning data centers won’t slow AI. His math — efficiency gains from software and hardware outpace new construction by about an order of magnitude, and one state’s moratorium costs the industry roughly 5-10 hours of catch-up — rests on the distinction that aggregate compute barely moves the industry’s overall progress while relative compute decides the competitive race, which is why labs fight for GPUs anyway. He redirects the backlash toward what works: the credible threat of bans forces companies to negotiate (communities should extract real investment), workplace-level decisions — “AI psychosis among CEOs” — are the better target, and “good old fashioned unionization” (Korean strikes, Hollywood’s AI protections) is the proven lever. The final HatGPT sends off the four-year segment with the week’s best items: Apple Vision Pro surgery (19% shorter tear-duct procedures), GTA 6 leaks including a nudist colony, an Amazon drone dropping a package in a Texas woman’s pool, carrier-pigeon texting apps, WSJ’s AI-agent insomniac founders, the “meat proxy” coinage for people who blindly relay AI output (“a kind of creeping human disempowerment”), Tiangong the robot sprinter doing 8.86 seconds then running straight into a padded wall, and Time’s “Inside OpenAI’s Reboot” — Mark Chen’s claim that OpenAI is 80% of the way to AGI, which Kevin counters with “by any pre-2022 definition of AGI, AGI is here.”
“Please do not wait until truly the last possible second to do the bare minimum, effectively at gunpoint from 47 attorneys general, and then say, we’re excited to announce our new leadership position in teen safety.” — Casey, on Meta’s settlement spin
“If one state stops new data center construction, that translates to something like 10 hours… that’s how much it takes aggregated for the AI industry to catch up through efficiency improvements to the compute that has been foregone.” — Arvind Narayanan
“While the total amount of compute is not a big factor in the aggregate rate of AI progress, the relative amount of compute is a surprisingly big factor in the relative competitive positions of companies. Both of those things are true at the same time.” — Arvind Narayanan
“Imagine fighting for your constitutional right to send a 13-year-old a push notification at 3 a.m. saying that there was a new reel that they might be interested in. At some fundamental level, that is what this case has been about.” — Kevin, on the Meta settlement