Meta on Trial + Is A.I. a ‘Normal’ Technology? + HatGPT

Episode Summary

Executive Summary: The episode centers on Meta’s antitrust trial, arguing that the FTC’s market definition may be outdated in a TikTok-shaped social media world, then shifts to a skeptical, nuanced discussion with Princeton’s Arvind Narayanan about why AI should be treated as a normal, slower-diffusing technology rather than imminent superintelligence. It closes with humorous Hat GPT commentary on tech news, from hacked crosswalks to A1/AI confusion.

Main Topics: Meta antitrust trial and market definition (Priority: 5/5): Kevin and Casey review the FTC’s case that Meta illegally maintained a monopoly in “personal social networking” by buying Instagram and WhatsApp, while Meta argues the market is fake and outdated. Zuckerberg’s internal strategy and Instagram spin-off ideas (Priority: 4/5): The hosts discuss newly surfaced emails showing Zuckerberg worried about Instagram’s “strategy tax,” considered spinning it off, and viewed it as potentially more valuable as an independent company. TikTok’s disruption of the social media market (Priority: 5/5): They argue Meta may once have had monopoly-like power, but TikTok’s emergence reshaped the market, making the FTC’s 2016-era theory harder to prove today. AI as normal technology vs. fast takeoff (Priority: 5/5): Arvind Narayanan and Sayash Kapoor’s paper argues AI is powerful but will diffuse slowly like electricity or PCs, with organizations and regulation acting as bottlenecks. AI safety, deployment, and accountability (Priority: 4/5): The discussion distinguishes capability from power, emphasizing that harms often arise at deployment, not just model development, and that deployers should share responsibility. Hat GPT roundup of tech absurdities (Priority: 3/5): A comedic segment covers hacked crosswalks mocking Zuckerberg/Musk, Cuomo’s ChatGPT-tinged housing plan, Dolphin Gemma, Linda McMahon calling AI “A1,” 3D-printed infrastructure, and Blue Origin’s all-women flight.

Key Arguments: The FTC’s case against Meta depends heavily on a narrow market definition, and that definition looks weaker now that social media has shifted toward algorithmic discovery and short-form video. Meta likely had stronger market power years ago, but the rise of TikTok means the alleged monopoly is not as clearly maintained today. Zuckerberg’s internal concern that Instagram was worth more as a standalone company is evidence that even Meta understood the strategic tension of the acquisition. AI is improving quickly, but adoption and organizational transformation are slower; widespread use does not equal immediate social upheaval. The biggest AI risks may come from deployment choices, weak governance, and misuse, not necessarily from a rogue superintelligence scenario. Market incentives and regulation can discourage unsafe autonomous AI deployment, especially if responsibility is clearly assigned. The debate over AI risk is distorted by hype on one side and dismissal on the other; the more useful position is nuanced skepticism. Some near-term AI harms, like deepfakes and misinformation, deserve more urgent policy attention than far-off superintelligence claims.

Data Points: FTC case filing date: December 2020 - The original Meta antitrust complaint was filed during the first Trump administration. Instagram acquisition year: 2012 - Meta bought Instagram before the current trial. WhatsApp acquisition year: 2014 - Meta bought WhatsApp two years after Instagram. Trial venue: U.S. District Court in Washington, D.C. - The Meta antitrust trial began there this week. Potential remedy sought: Divest Instagram and WhatsApp - The FTC is seeking a breakup remedy that would undo major Meta acquisitions. Instagram’s share of revenue: Around half of Meta’s overall revenue - Used to illustrate how central Instagram has become to Meta’s business. ChatGPT user base: Something like 500 million users - Cited during the AI diffusion discussion to show rapid adoption. U.S. adults using generative AI: About 40% - Referenced as evidence of widespread but not necessarily intensive use. AI work usage intensity: About one hour per work week - Narayanan cited this to argue usage is still limited in practical terms. Productivity gain estimate: A fraction of a percentage point - Narayanan described the effect of current AI use on productivity as very small. Meta settlement offer: $450 million - Reportedly offered to settle the FTC case, far below the government’s demand. FTC demand: $30 billion - The FTC’s asking price in the settlement negotiations. Meta payments to Trump-related entities: $26 million total - Discussed as part of Meta’s lobbying/appeasement efforts. Inauguration contribution: $1 million - Part of the reported $26 million paid by Meta. January 6 lawsuit settlement: $25 million - Part of the reported $26 million paid by Meta. New 3D-printed train station build time: 6 hours - A Japanese station was assembled very quickly using printed components. New train station component prep time: 7 days - The station’s components were 3D-printed off-site over a week. New train station size: Just over 100 square feet - The new rural Japanese station is extremely small.

Pivotal Quotes: "It’s your body to rest, to nourish, to grow. It’s your mind, you know. It’s your place, your life, to love, to dream, to change." — Host readout / sponsor copy: Opening New York Times promo framing the episode’s theme of personal agency. "The market that the government is suggesting they have a monopoly over is fake and has been sort of invented solely for the purposes of this trial." — Casey Newton: Summarizing Meta’s defense against the FTC’s market definition. "For us, a lot of the safety concerns come from the deployment phase as opposed to the development phase." — Arvind Narayanan: Explaining why his framework focuses on how AI is used, not just how powerful models become.

Implications: The episode suggests antitrust law struggles to keep up with fast-changing tech markets, while AI policy should focus less on apocalypse narratives and more on practical governance, deployment controls, and near-term harms like deepfakes, surveillance, and misuse.