Summary

Social platforms have turned rage into a scalable political asset, rewarding content that approaches a rule without quite breaking it. Kevin Roose argues Charlie Kirk mastered this “borderline content,” while Casey Newton traces how both right- and left-wing creators adopted the language of “destroys,” “owns,” and “humiliates.” The result is a culture “optimizing for rage,” with few stabilizing forces willing to sacrifice engagement.
X is the clearest commercial proof that escalating political temperature can still drive growth. After Kirk’s assassination, Elon Musk declared, “The left is the party of murder,” while X recorded more first-time US downloads than on any previous day, according to Nikita Bier, who works in product at X. Roose stresses X’s unusual elite influence, though Newton pushes back that rage bait now defines Bluesky, TikTok, Instagram, and the broader app-based media ecosystem too. Roose also notes that, a week after the shooting, there had not been a huge nationwide upsurge of violence—evidence that most Americans do not want a violent cultural war, despite what the feeds suggest.
The hosts see technological off-ramps, but today’s business incentives point the other way. “Bridging-based algorithms” could rank material that earns agreement across political groups, extending the logic behind Community Notes into core feeds. Yet Facebook’s partly unimplemented 2020 “break-glass” measures showed the trade-off plainly: lowering hostility is possible, but “it was bad for engagement.”
The episode also warns that rage is becoming a tool of official and crowdsourced punishment. Government pressure on broadcasters over political speech, along with organized campaigns to report tasteless jokes to employers, threatens the First Amendment’s protection for offensive expression and gives ordinary users incentives to police one another.
A $2 billion UAE transaction with World Liberty Financial coincided with a US policy opening for advanced Nvidia-chip exports, creating extraordinary conflicts without proving a quid pro quo. MGX used World Liberty’s stablecoin for a $2 billion Binance investment, potentially generating tens of millions of dollars for the Trump- and Witkoff-linked company. Around the same period, the administration advanced a tentative chip agreement sought by the UAE despite concerns that technology could reach China; David Yaffe-Bellany’s careful conclusion is overlapping people, timing, and interests—not evidence of an explicit quid pro quo.
Crypto has become a global, frictionless influence channel around the Trump administration. Yaffe-Bellany calls it a recurring pattern: parties seeking policy outcomes increasingly consider investing in World Liberty tokens or joining Trump-linked crypto events, even when no direct exchange can be demonstrated. Newton’s analogy is “the Trump Hotel on steroids”—foreign buyers can deploy billions online despite restrictions on foreign campaign contributions.
Publishers face simultaneous pressure to automate their journalism and defend it from AI intermediaries. Business Insider authorized AI-written first drafts while warning reporters to “make sure your final work is yours,” a policy Roose calls “playing with fire” because it narrows the distance between assistance and job replacement. Separately, Penske says Google’s AI Overviews are suppressing traffic; its affiliate revenue was down more than one-third from its peak by the end of last year.
Consumer AI is broadening rapidly while ChatGPT and Claude appear to be separating into distinct franchises. OpenAI says 52% of ChatGPT users had typically feminine names by July 2025, usage was growing fastest in lower-income countries, and about 73% of messages were non-work-related; “practical guidance” ranked first and information-seeking second. Anthropic instead found Claude adoption led by coding and business automation, supporting Roose’s thesis that ChatGPT is becoming ubiquitous while Claude becomes “a thing that you use if you’re building something with code.”
AI safety measures for teenagers may reduce harm on one platform while displacing vulnerable users elsewhere. OpenAI plans to infer whether users are under 18 and provide a restricted ChatGPT that blocks graphic sexual content and may, in rare cases, involve law enforcement. Roose welcomes the move but warns that teens could migrate to less-protected services or private local models: “The horses are sort of out of the barn here.”

Deep dive

1. Borderline content converted outrage into political power

Kevin Roose’s framing: Kirk excelled at claims that approached platform-policy boundaries without clearly crossing them, including unsupported vaccine-death speculation, criticism of the Civil Rights Act, and attacks on Jewish donors followed by denials of antisemitism. “The most compelling thing you can do on social media is to almost break a policy.”
A graph Mark Zuckerberg once shared captured the mechanism: wherever a platform draws its enforcement line, engagement rises as content gets closer to it. Facebook and YouTube promised to down-rank such material, yet creators continually discovered new boundaries to exploit.
Roose’s milk-crate-challenge example carries the broader point: TikTok lacked a rule until videos showed people falling from dangerously stacked crates. Systems selecting “the most engaging thing” will repeatedly surface behavior “just on the brink of maybe not being okay.”
Newton adds the escalation loop: Kirk’s “debate me bro” clips used violent verbs such as “destroys,” “humiliates,” and “owns”; left-wing creators copied the aesthetic because calmer material received little distribution. “We as a culture are optimizing for rage now.”

2. X intensifies the conflict, but rage bait has escaped every platform

Roose contrasts X with old Twitter, whose teams sometimes removed violent footage, hid it behind warnings, down-ranked it, and banned persistently inflammatory accounts. Elon Musk restored many of those accounts, relaxed restrictions, and used the platform’s largest audience to promote his own views.
Immediately after the assassination, before the shooter’s motives were established, Musk posted, “The left is the party of murder.” For elites still checking X “20, 100 times a day,” Roose argues, the feed can make imminent civil war appear more representative of America than it is.
Newton’s pushback—worth keeping: the temperature is also rising on Bluesky, TikTok, and Instagram; “I do not feel like there is a place on the internet right now that is being optimized for civil and respectful discussion.” Roose agrees, while saying he has seen less violent left-wing material on Bluesky than on X.
The uncomfortable business signal: X recorded more first-time US downloads that week than on any day in its history, according to Nikita Bier. However “violent,” “awful,” and “rage-filled” the experience became, Roose notes, it appeared to work by that shallow growth metric.

3. Platforms possess de-escalation tools but lack the incentive to use them

Facebook developed “break glass” measures for the 2020 election—internal levers intended to lower News Feed hostility if the country seemed to be spiraling toward violence. Not all were deployed, and the company quickly encountered the governing constraint: reducing discord hurt engagement.
Roose’s proposed alternative is Aviv Ovadya’s “bridging-based algorithms.” Community Notes already surfaces corrections when differently aligned users agree; applying that consensus mechanism to core-feed ranking might reward material that connects political groups rather than material that divides them.
Roose admits the commercial uncertainty: a bridging feed “may be less engaging and a worse business.” Still, the existence of cross-partisan ranking on X and Meta demonstrates that technology can seek consensus even “on some of the worst platforms on Earth.”
Newton avoids technological determinism: the hosts do not know the assassination’s full motives, and “nothing is ever just the Internet’s fault.” His narrower concern is that society has absorbed platform logic even where no algorithm directly dictates an individual action.
A week after the shooting, Roose notes there had been no huge nationwide upsurge of violence; he takes that as evidence that most Americans do not want a violent cultural war. Both hosts urge a calmer media diet rather than letting platforms’ civil-war image become reality.

4. Political speech faces government pressure and crowdsourced punishment

The hosts point to FCC chairman Brendan Carr’s criticism of Jimmy Kimmel’s remarks and ABC’s subsequent indefinite leave for the comedian. Roose calls government threats against broadcasters over political speech “terrifying, full stop”—precisely the offensive or tasteless expression the First Amendment is meant to protect.
Newton sharpens the democratic test: “You really do not wanna live in a country where the comedians on TV cannot make fun of the president.” Roose contrasts that pressure with years of Republican and Kirk-led campaigning for broader conservative speech rights on platforms.
Newton also describes an organized “snitching culture”: users submitted tasteless posts by neighbors, Facebook friends, and small-business owners to amplifiers and databases, then contacted employers seeking firings. The behavior resembles Gamergate-era campaigns, but he had not seen it operate at this scale.
Roose calls doxing “a kind of video game that you can play on X,” made exciting by its consequences for real lives. His uncomfortable concession is that free speech also permits demands that someone lose a job; protecting political expression therefore requires accepting some ugly downstream trade-offs.

5. The UAE’s $2 billion crypto deal and chip breakthrough share a cast

Yaffe-Bellany’s investigation begins with Sheikh Tahnoon bin Zayed Al Nahyan and Middle East envoy Steve Witkoff meeting on a superyacht off Sardinia. The White House described conflict-resolution diplomacy, but the reporters found a relationship spanning government matters and business ties between their family-linked organizations.
In the first deal, UAE investment firm MGX used World Liberty Financial’s stablecoin to make a $2 billion investment in Binance. Because stablecoin issuers earn yield on reserves, that large principal could generate tens of millions of dollars for the Trump- and Witkoff-linked company.
The second deal is a tentative US-UAE agreement permitting exports of powerful American Nvidia chips. The UAE had pursued them for years, while US officials worried that close Chinese ties—including joint military exercises and technology-sharing agreements—could let the chips or capabilities reach China.
Newton summarizes the apparent sequence bluntly: the UAE wanted chips, directed $2 billion through the Trump family’s crypto business, and then saw US policy soften. Yaffe-Bellany’s correction is essential: “We did not find evidence of a kind of direct quid pro quo,” only intertwined timelines, personnel, and interests.

6. Personal financial interests blurred into official chip policy

One concrete overlap makes the concern harder to dismiss: an employee of G42, Sheikh Tahnoon’s chip-seeking Emirati technology firm, simultaneously worked for World Liberty while both deals developed. Yaffe-Bellany says this raises ethics and conflict questions without proving US policy was purchased.
Witkoff remained financially interested in World Liberty while serving as Trump’s Middle East envoy, despite earlier statements that he would divest. His disclosure appeared shortly before the investigation; the administration said he remained “in the process of divesting” eight months into its term.
AI and crypto czar David Sacks aggressively supported the chip agreement while continuing work at Craft Ventures. Craft had invested in AI, and one original backer was an Emirati firm chaired by Sheikh Tahnoon; White House ethics waivers deemed Sacks’s remaining interests de minimis, though colleagues still raised concerns.
David Feith, the National Security Council’s export-control advocate, was removed after Laura Loomer urged Trump to fire him and colleagues. His departure eliminated a potential obstacle and let Sacks take greater control, though Yaffe-Bellany repeatedly hedges that the reporting cannot establish every motivation behind closed-door decisions.

7. The chip-policy flip replaced containment with ecosystem expansion

Newton had watched a China-hawkish administration reverse course surprisingly quickly. The earlier premise was straightforward: if advanced AI depends on Nvidia chips, a country racing China should prevent its adversary from acquiring them.
The replacement doctrine, promoted by Sacks and others, held that broader distribution was good because foreign companies and governments would build on American technology. The investigation helped Newton connect that intellectual shift with the UAE negotiations, personnel changes, and overlapping financial relationships.
Yaffe-Bellany stops short of claiming the deals caused the policy reversal: the project established “the timelines of how these two deals developed and how they intersected.” He expects more facts to emerge because the reporting was unusually difficult and distributed across sources on three continents.

8. Crypto functions like a global influence market without proving explicit trades

Reporting began when Zach Witkoff announced the $2 billion transaction at a Dubai crypto conference in late April or early May. Yaffe-Bellany’s immediate question was how a foreign leader came to channel money toward a top White House adviser’s family; Zach suggested Americans should “take a page out of the Emirati royal’s book.”
The White House and World Liberty denied any connection between the private crypto deal and government chip policy. Officials said Sacks held no financial stake in the chip agreement, did not previously know its key Emirati players, and pursued administration priorities; G42 cited conflict-protection protocols.
Yaffe-Bellany nevertheless sees a crypto-industry pattern: actors seeking regulation or access increasingly consider World Liberty investments, Trump tokens, or the meme-coin dinner. None establishes an explicit quid pro quo, but collaboration with “the Trump family apparatus” is becoming part of how business gets done.
Newton calls crypto “the Trump Hotel on steroids”: purchases are global, instant, and capable of reaching $2 billion, while foreign buyers barred from campaign donations can still support the president’s businesses. Ethics groups object, but earlier scrutiny only delayed stablecoin legislation; it ultimately passed with Democratic votes.

9. AI is squeezing publishers from both sides

Business Insider told reporters they may use AI for first drafts if they “make sure your final work is yours.” Roose already uses AI to catch typos and possible factual errors, but writing crosses his current boundary and risks helping “automation away” the reporter’s own job.
Newton sees an immediate reputational hazard because readers can sometimes recognize AI’s stylistic flaws and may lose respect for a publication. Longer term, if machine writing becomes indistinguishable, he expects it to disappear into “the background noise of the internet.”
The contradiction is unusually sharp: Business Insider and other outlets previously removed fabricated stories allegedly submitted by an AI-using freelancer, yet Business Insider now permits staff-generated AI drafts. Roose’s verdict: “We’re just in stupid town right now.”
Penske’s lawsuit attacks the distribution side, alleging Google’s AI Overviews reuse reporting while depressing traffic. Penske says affiliate-link revenue was down over one-third from its peak by the end of last year; Roose will not declare Google’s conduct illegal, but says the publisher-level economics are “bad for the overall health of the web.”

10. Consumer AI expands as safety and product strategies diverge

Albania’s AI minister Diella is supposed to make public tenders “100% free of corruption.” The hosts see AI-assisted government as inevitable but doubt current systems are ready to monitor corruption or directly exercise power.
Roblox game Steal a Brainrot reportedly reached 23.4 million concurrent players, translating TikTok’s AI-generated Italian Brainrot characters into a massive teen-facing phenomenon. Its copyright irony arrived when Tung Tung Tung Sahur disappeared during licensing discussions: someone effectively stole a character from the game about stealing Brainrot.
OpenAI’s planned teen experience will infer age from conversations, block graphic sexual material, and sometimes involve law enforcement. Newton calls it overdue amid an FTC inquiry and a lawsuit from the family of Adam Raine, a 16-year-old who died by suicide; Roose warns privacy-conscious teens may simply migrate elsewhere.
OpenAI’s consumer data shows 52% of users had typically feminine names by July 2025, about 73% of messages were non-work-related, and practical guidance outranked information-seeking. Anthropic’s consumer-plus-API study instead found coding and automation dominance, with highest per-capita Claude use in Singapore and Israel; within the US, the leading states or jurisdictions were Washington, DC, and Utah—evidence supporting Roose’s thesis of increasingly distinct mass-market and enterprise trajectories.

Deep Learning is Not So Mysterious or Different - Prof. Andrew Gordon Wilson (NYU)Vol.183 Industry Watch 33 | The Power Struggle in the AI Era: Reframing China’s New Power System