Summary

Prediction markets have reached a scale where privileged information is becoming the edge, undermining the claim that prices represent collective intelligence. An Army sergeant allegedly connected to Nicolás Maduro’s capture made more than $400,000, while analysis of 400,000 Polymarket markets found military and defense long shots won 52% of the time versus a 14% platform average. Casey Newton’s verdict: without inside information, “you kind of have to be a sucker to participate.”
The consumer economics already resemble gambling more than reliable information production. More than 70% of Polymarket users lose money, and Kalshi recently had 2.9 unprofitable users for every profitable one; suspicious Super Bowl and temperature bets further weaken confidence. Market integrity is not cosmetic: insider advantages destroy the trust required for markets to remain “liquid and transparent.”
Regulatory pressure is rising, but the CFTC’s small size and jurisdictional fight leave enforcement thin. Proposed measures range from age verification, self-exclusion and advertising limits to surveillance of insider trading; the Senate barred senators from betting, while a Gillibrand-McCormick bill would cover legislative and executive officials. Brazil blocked 27 sites, including Kalshi and Polymarket, and France and Hungary imposed bans, while Kevin Roose put “high probability mass” on U.S. action against flagrant military or congressional abuses by year-end.
Joanna Stern’s year of AI experiments found agents and wearables progressing much faster than humanoid robots. A reporting assistant’s research and email workload went from requiring a human early in the year to, in Stern’s assessment, being 100% automatable now; persistent assistants across the Bee bracelet and Meta glasses also felt increasingly plausible. Robots coming to live with consumers, however, are “really not coming to live with us anytime soon.”
Dental AI supplied the episode’s clearest warning about monetized machine judgment. An AI overlay helped a dentist recommend four sessions of periodontal treatment costing potentially thousands, but other dentists found that better home care was sufficient. Staff later told Stern that practice owners used AI reports to question why clinicians had not drilled or sold treatments—the “fancy high-tech sheen” can turn probabilistic detection into an upsell engine.
Stern’s most useful AI was a mirror and editor, not an autonomous author or trustworthy authority. ChatGPT wrongly declared her dying praying mantis pregnant, yet helped her decide to leave The Wall Street Journal after processing her plans, projections and fears: “It kind of did tell me what I wanted.” She wrote every word of her book herself, using AI for editing, fact-checking and endnotes while relying on humans for long-form structure, illustration and fact-checking.
Attention School treats degraded attention as a collective political problem, not merely a screen-time habit. Its exercises restore perception, play and shared physical experience, while its founders cast every uncommodified moment as resistance to the “fracking of our eyeballs.” Rachel Cohn reported no transformation after one month, but found an unusually engaged community of ordinary technology users seeking alternatives before AI further reshapes daily life.

Deep dive

1. Prediction markets are rewarding access before insight

Casey Newton calls prediction markets one of the year’s defining technology stories: once a niche academic interest among Bay Area “absolute nerds,” they now occupy the popular imagination and advertise across New York. Kevin Roose’s early fear—that legalization would produce “a total casino”—looks less theoretical now.
An Army sergeant allegedly involved in Nicolás Maduro’s capture made more than $400,000 betting that Maduro would be out of power by the end of January. The Anti-Corruption Data Collective’s broader finding is harder to dismiss: across more than 400,000 settled Polymarket markets over five years, military and defense long shots won about 52%, versus 14% for the platform overall.
Kevin Roose’s Strava analogy captures the escalation: military activity was once exposed inadvertently through exercise heat maps; now personnel may be directly monetizing operations they participate in. Newton’s sardonic incentive test: why “collect your freaking paycheck like a chump” when privileged information can be wagered online?

2. Manipulation and misinformation corrode the product together

At Paris Charles de Gaulle Airport, the recorded temperature jumped from 18°C to 22°C on April 15 amid suspiciously timed Polymarket bets and an allegation that measuring equipment was interfered with. How it was tampered with remains unknown.
The photograph seemingly showing someone heating the sensor with a hairdryer was itself AI-generated and circulated in a prediction-market Discord. Roose’s correction matters: the episode is simultaneously about possible market manipulation and “slop and disinformation”—even the supposed evidence of cheating was synthetic.
Super Bowl markets on Bad Bunny’s songs and celebrity guests created another obvious information asymmetry: performers, rehearsal attendees or entourage members could trade against outsiders. After enough such incidents, Newton argues, ordinary participants will conclude that “you kind of have to be a sucker” to bet without inside information.

3. The house is winning while trust is disappearing

More than 70% of Polymarket users lose money, according to reporting cited by Roose. At Kalshi, recent data showed 2.9 unprofitable users for every profitable user—the relevant baseline for anyone seeing ubiquitous ads and imagining “a quick buck.”
Roose links those losses to the purpose of insider-trading law: the harm is not merely transferring money from one participant to another. Persistent unfairness destroys the confidence that lets a market remain “liquid and transparent,” eventually driving away recreational users and weakening prices as information signals.
That creates a direct conflict with the industry’s pitch. These markets are supposed to discover “the true price of things” through collective wisdom, yet their most conspicuous winners increasingly appear to be people who can alter an outcome or know it in advance.

4. The regulatory gap is structural, not accidental noise

States have tried to ban prediction markets, but the CFTC has sued to defend its exclusive jurisdiction. The hosts’ framing is stark: the federal regulator says the field is its domain while showing limited appetite and capacity to police it.
Roose says the CFTC is only a fraction of the SEC’s enforcement scale and inherited prediction markets through a “historical accident”: Kalshi’s products qualified as futures contracts. Newton would not be surprised if platforms lobbied to remain there, comparing the preference with crypto firms wanting the CFTC rather than an SEC that is “really good at their jobs.”
The Senate unanimously barred senators from prediction-market betting. Senators Kirsten Gillibrand and Dave McCormick also introduced a bill covering legislative and executive officials; Roose immediately asked whether Senate staff could do it, leaving other privileged actors as an obvious question.
Brazil blocked 27 sites, including Kalshi and Polymarket, as illegal gambling; France and Hungary also imposed bans. America’s contrasting posture, in Newton’s caricature: “For there is money to be made. Go forth and make it.”

5. Useful markets require both gambling controls and enforcement

One early prediction-market originator argued that insider trading improved prices: Bad Bunny’s entourage or military personnel could reveal information society would otherwise lack. Roose calls that “a beautiful theoretical construct” with “zero chance of surviving contact with the real world”; Newton says the actual incentive is to betray friends, family, coworkers and country.
Roose separates two harm classes. Gambling controls should include self-exclusion, mandatory age verification and advertising limits, preventing a future in which Kalshi becomes “the hottest thing” in high schools and 16-year-olds accumulate debt betting on the Super Bowl.
Newton says market integrity needs a “big, bad regulator” actively surveilling trades and removing bad actors. He argues Kalshi and Polymarket should welcome that oversight because their prices might then become useful rather than reflecting whoever can manipulate a sensor or exploit confidential access.
Roose still wants the original knowledge-production vision: monetary incentives could fund independent polling and research better than today’s institutional model. By year-end, he assigns high odds to rules targeting blatant congressional and military abuses, especially because wagering on overseas operations creates a national-security problem.

6. Stern’s experiment separates near-term tools from theatrical promises

Joanna Stern built I Am Not a Robot around a simple challenge to executive rhetoric: if AI will change jobs, healthcare, transportation and “the fabric of our lives,” she would test those claims across an entire year. The result is deliberately time-bound—a snapshot readers can revisit in five or 10 years to see where she was “totally right” or wrong.
Her experiment ranged from an AI companion named Casey, described as shallow and full of “robo-horniness,” to Waymos, customer support, medicine, parenting, meal planning and book production. The point was not that every tool worked, but that the present contains things “clearly hype,” “sometimes quite good” and sometimes “quite terrible.”
Humanoid robots landed firmly in the hype category for domestic use: “They’re really not coming to live with us anytime soon.” Stern nevertheless finds their training compelling and dystopian—machines must observe humans folding laundry, washing dishes and performing other mundane physical tasks.

7. Agents and persistent wearables moved fastest during the year

At the year’s start, Stern hired a reporting assistant for research and email tasks. By midyear, Perplexity Comment was performing much of that work reasonably well; by the interview, she judged that an agent could do “100% of those tasks.”
Wearables also surprised her. No single device completed the vision, but elements from the Bee bracelet, Meta glasses and other products suggested an AI assistant could persist throughout the day on something worn rather than opened on a computer.
That persistence already raised a social boundary: the bracelet’s apparent recording and transcription prompted Stern’s Wall Street Journal colleagues to insist, “Please leave your bracelet at the door.” Her boss repeatedly told her, “Do not wear that in here”—a useful boundary on ambient assistants before their technical promise is mistaken for permission.
Parenting supplied a cleaner warning. When Stern’s son’s praying mantis turned brown, ChatGPT live mode enthusiastically announced it was pregnant; the mantis was dying. For her four- and eight-year-olds, AI literacy therefore starts with exposure plus the recurring lesson that the system can be “fully wrong.”

8. Dental AI turns detection into a sales incentive

A dentist showed Stern a Pearl AI overlay that placed vivid boxes around cavities and highlighted plaque buildup, then recommended deep cleaning and periodontal treatment across four sessions. The work might not be covered by insurance and could cost thousands of dollars, despite Stern having no comparable history or troubling symptoms.
Other dentists saw the same AI reading but judged the condition “really not that bad” and recommended better home care. Stern never received the periodontal treatment, making the disagreement more consequential than a routine false alarm.
Anonymous dental-office workers then described the organizational mechanism: dental service organizations could inspect AI reports and ask clinicians why a cavity was not drilled or periodontal treatment was not sold. The system did not merely support diagnosis; it gave owners a standardized dashboard for pressuring treatment volume.
Stern preserves the medical nuance. Detecting tiny abnormalities can be valuable in breast-cancer screening, especially given her family risk, but more sensitivity is not automatically better in every context. Roose’s darker formulation: AI’s authority can make an unnecessary service feel like something “a human would have missed.”

9. The best chatbot advice reflected Stern’s own accumulated evidence

When Stern considered leaving The Wall Street Journal after 12 years, colleagues hedged; ChatGPT said, “I think you should go. You should quit.” She had supplied notes, financial projections, fears and risk-reduction plans, letting the system organize evidence she felt too anxious to interpret.
Her conclusion is deliberately double-edged: AI is “this mirror,” and “it kind of did tell me what I wanted.” The decision has worked so far, but she concedes, “Had it not, I would say this stuff is stupid”—a reminder that perceived wisdom is often judged retrospectively by outcome.
Stern wrote every word of the book, using AI for copy editing, fact-checking assistance and an endnotes process she says would otherwise have been impossible. A human editor repaired long-form structure after the model responded, “This is great. This is the best book I’ve ever read”; humans also handled illustration and fact-checking.
She resists a simple gender narrative despite figures showing men 22% more likely to be heavy workplace users and 61% of women expecting more harm than good. Industry composition partly explains adoption, while Stern sees the age divide as more urgent: younger workers blame AI for scarce jobs, although she stresses that causation remains unclear.

10. Attention School rejects productivity as attention’s only purpose

Brooklyn’s Struthers School of Radical Attention serves people from 7 to 70, though evening and weekend programs largely resemble adult continuing education. Most offerings are free; the seminar Rachel Cohn attended cost The Times $250.
The school does not prescribe phone abstinence or treat technology as an enemy. It asks participants to study and practice attention beyond narrow focus and productivity, while confronting systemic harms and the “commodification of our attention.”
In one paired exercise, one person could speak but not question, while the listener could ask questions but not make affirmative statements. Even as a professional interviewer, Cohn found the constraint awkward and clunky—which was the point: it exposed habits normally hidden inside conversation.
A Georges Perec-inspired exercise asked participants to “exhaust” a neighborhood space through observation. Cohn recorded Sweetgreen workers, trash and passing pant legs; when participants shared one line each, they constructed a collective place and revealed perceptual differences, including one woman’s realization that she attended intensely to sound rather than sight.

11. Attention becomes resistance when unmonetized life forms community

The school’s movement rests on study, sanctuary and coalition building, but Cohn repeatedly pressed for concrete political objectives. Co-founder Peter Schmidt’s answer challenged the premise: politics need not begin with policy when gathering to surf, observe or talk spends time that technology companies cannot monetize.
A sidewalk study made that theory bodily. After reading Anthony Bourdain’s contrast between the body as temple and “an amusement park ride,” participants explored a farmers market, then shared oysters, focaccia and perceptions—an exercise Roose likened to reintroducing a cloud-uploaded mind to lettuce and strawberries.
In a paid radical-imagination seminar, participants identified the “prison guard” constraining their imagination, then created characters embodying qualities they wanted to expand. Cohn revived her six-year-old alter ego, Princess Lollipop, after realizing that rigidity and impatience were preventing her from approaching the program playfully.
Roose connects the school to Buddhism, improv and earlier countercultures such as the Transcendentalists’ response to industrialization. Newton adds Silicon Valley’s former countercultural roots and the newer counterculture rejecting its dominance. Its “Friends of Attention” deliberately borrow environmental language—“re-enchanted with nature” and the “fracking of our eyeballs”—to describe extraction and repair.
Cohn’s honest assessment: one month produced no transformational breakthrough, only gradual insights resembling group therapy. Yet the engaged mix of scientists, civil servants, knowledge workers and a minority of neo-Luddites convinced her that people want a place to ask what a flourishing human life means while continuing, in most cases, to use technology.

Ben Horowitz on American Dynamism and the Future of AI | The a16z Show How Kaz Nejatian Is Rebuilding Opendoor