E10: Twitter & Facebook botch censorship (again), the publisher vs. distributor debate & more

Summary Chamath predicted that after Amy Coney Barrett’s quick confirmation, a Clarence‑Thomas‑style challenge to Section 230 would be fast‑tracked to the Supreme Court and, as a result, Section 230’s current liability shield would effectively be ended or fundamentally overturned “within the subsequent few years.” That has not happened as of November 29, 2025.

What actually happened

1. Amy Coney Barrett was confirmed quickly, as assumed.
   The Senate confirmed Barrett to the Supreme Court on October 26, 2020, just days after the podcast, matching the predicate of his forecast. (en.wikipedia.org)

2. A major Section 230–related case did reach the Supreme Court, but the Court declined to change the law.
   Gonzalez v. Google LLC (about whether Section 230 immunity covers recommendation algorithms) and Twitter, Inc. v. Taamneh were heard in February 2023 and decided in May 2023. The Court unanimously ruled for the tech companies in Taamneh and then vacated and remanded Gonzalez without deciding the scope of Section 230, explicitly “declin[ing] to address” Section 230 and leaving its broad liability shield intact. (en.wikipedia.org)
   This is the opposite of the “fundamental overturning” Chamath envisioned.

3. Subsequent Supreme Court actions have reinforced, not dismantled, Section 230.
   In October 2025, the Supreme Court refused to hear an appeal in a case where a plaintiff alleged Grindr’s design enabled his sexual assault; the Ninth Circuit had held that Section 230 barred the suit. By denying certiorari, the Court let that broad Section 230 immunity ruling stand. (reuters.com)

4. Section 230 remains the baseline federal liability shield as of late 2025.
   Authoritative overviews of Section 230 note that, after Gonzalez/Taamneh, the Supreme Court “effectively avoided” changing Section 230, and the statute continues to provide broad immunity to platforms for user‑generated content. (en.wikipedia.org)
   Some lower‑court cases (like Anderson v. TikTok in 2024) have narrowed immunity at the margins (e.g., for certain algorithmic “recommendations”), and Congress has repeatedly introduced—but not passed—bills to amend or sunset Section 230 (EARN IT Act, Algorithm Accountability proposals, etc.). Those efforts show political pressure but do not amount to Section 230’s liability regime being “ended” or fundamentally overturned. (en.wikipedia.org)

Evaluation of the prediction

• Correct components: Barrett was indeed confirmed within days, and significant Section 230–related cases did reach the Supreme Court relatively soon afterward.
• Crucial failure: The key claim—that these developments would lead the Court to effectively end or fundamentally overturn Section 230’s liability shield “within the subsequent few years”—is clearly contradicted by the Court’s 2023 decisions (which sidestepped changing Section 230) and its 2025 refusal to take up another challenge to those protections.

Because the core outcome he “bet” on (Section 230’s days being “numbered”) has not materialized more than five years later and the law’s central immunity remains intact, the prediction is wrong.

There is no clear empirical test of Chamath’s quantitative claim, so its correctness can’t be determined from available evidence.

• No major ad‑supported social network has actually done what the prediction describes. Since 2020, Facebook, Instagram, and X/Twitter have all kept algorithmic feeds as the default, while offering optional reverse‑chronological modes (e.g., Facebook/Instagram’s “Most Recent” / “Following,” X/Twitter’s “Following” vs. “For You”). None has permanently replaced the algorithmic feed with a purely reverse‑chronological feed for all users, which is the scenario the prediction is about. (wired.com) TikTok remains entirely algorithmic, and alternative platforms that are chronological by default (e.g., Mastodon) are not large, ad‑driven networks comparable to Facebook or Twitter. (en.wikipedia.org)

• Internal and academic experiments compare algorithmic vs. chronological feeds but don’t publish the key metric Chamath specifies (revenue per page/impression). Twitter ran a large randomized experiment committing nearly 2 million users to a reverse‑chronological feed to study political amplification, not monetization. (arxiv.org) Meta’s 2020 election‑period experiment forced some users on Facebook and Instagram into chronological feeds for months; results published in Science and summarized by Wired show users on chronological feeds encountered more political and untrustworthy content and spent less time on Meta apps, often shifting that time to TikTok/YouTube. (wired.com) Earlier internal Facebook tests in 2014 and 2018 similarly reported lower engagement when ranking was removed, but again, not a quantified 90% drop in per‑impression revenue. (wired.com)

• One leaked Facebook test even suggests ad impressions (and thus total ad revenue) increased when feeds were chronological‑only, because people initially scrolled more and saw more ads, according to whistleblower‑derived reporting cited by the Chamber of Progress. (progresschamber.org) That cuts against the idea of an automatic catastrophic revenue collapse, but it still doesn’t tell us whether revenue per page/per impression fell by anything close to 90%.

Because (1) no major platform has fully and durably made the switch described, and (2) the experiments that do exist don’t disclose the specific monetization metric Chamath predicts (90% per‑page/per‑impression decline), the prediction remains untestable with public data. Hence it must be classified as ambiguous, not clearly right or wrong.

Why this prediction was wrong

1. The pandemic did not wind down in the ensuing months after October 2020. In the U.S., the deadliest wave up to that point hit immediately afterward: CDC data show the 7‑day average of new COVID‑19 deaths peaked on January 11, 2021, and by March 17, 2021, the U.S. had recorded over 535,000 deaths. (archive.cdc.gov) Globally, cases and deaths continued to climb through 2021, reaching about 292 million reported cases and 5.43 million deaths by January 1, 2022, and more than 652 million cases and 6.7 million deaths by late December 2022. (emro.who.int) The Omicron wave (Dec 2021–Feb 2022) then produced the largest U.S. case surge, with ~30 million cases and ~170,000 deaths in just three months. (mdpi.com)

2. The crisis phase lasted years, not months. The World Health Organization did not declare the end of COVID‑19 as a Public Health Emergency of International Concern until May 5, 2023—over 2.5 years after the October 2020 podcast. (who.int) Even after the emergency status ended, WHO and regional offices emphasized that COVID‑19 remained a significant ongoing health burden, requiring long‑term management. (paho.org)

3. Remdesivir and monoclonal antibodies did not dramatically end the crisis. WHO’s large Solidarity trial and related analyses concluded that remdesivir has little or no effect on overall COVID‑19 mortality, contradicting hopes it would substantially change outcomes. (who.int) While some monoclonal antibodies helped in early variants, many lost effectiveness as the virus evolved; by late 2024 the FDA had fully revoked emergency authorizations for several major COVID antibody products because circulating variants were resistant and the drugs were no longer in use. (reuters.com)

Overall, the COVID‑19 crisis intensified after mid‑October 2020 and only gradually shifted into a more endemic, manageable phase over several years, driven largely by vaccination and accumulated immunity rather than a rapid wind‑down over the “ensuing months” via remdesivir and monoclonal antibodies.

Contemporaneous coverage of NBC’s October 15, 2020 Trump town hall consistently described moderator Savannah Guthrie’s approach as unusually tough and adversarial, matching the prediction that NBC would “come out swinging.”

• Multiple mainstream outlets called it one of Trump’s toughest grillings as president. The Washington Post wrote that fears of a free pro‑Trump infomercial were unfounded because it became “one of the toughest grillings he has faced as president,” with Guthrie repeatedly challenging his evasions on topics like white supremacy, COVID‑19 deaths, and taxes.(washingtonpost.com) CBS News similarly said Guthrie gave Trump “one of the most aggressive interviews he’s experienced as president.”(cbsnews.com) CNBC’s write‑up described an “intense line of questioning” and noted Guthrie “grilling” Trump from the outset.(cnbc.com) The Guardian emphasized that she fact‑checked him in real time and “pushed back” on his rhetoric in ways he rarely faced.(theguardian.com)

• Conservative and pro‑Trump commentators also framed the town hall as an aggressive attempt to take him down, reinforcing that the tone was perceived as combative. Sean Hannity opened his show by calling it an “ambush,” saying Trump had effectively debated Guthrie, whom he labeled “Joe Biden’s surrogate.”(thewrap.com) Other Fox and right‑leaning figures said she was “badgering,” “embarrassing,” and acting as a “debate opponent,” and compared NBC’s town hall to an “interrogation” versus ABC’s “picnic” with Biden.(thedailybeast.com)

• Coverage explicitly linked Guthrie’s tough performance to NBC’s need to offset or atone for the controversial decision to give Trump a simultaneous prime‑time platform opposite Biden. NBC had been heavily criticized, including by its own staff, for scheduling the dueling town hall.(journalism.arizona.edu) Post‑event analysis noted that Guthrie’s “grilling” helped “take the heat off NBC for its town hall” and that she served as the face of a network decision many critics thought was not in the public interest.(journalism.arizona.edu) This is broadly in line with the user’s phrasing that NBC would try to “take down Trump” as a form of “penance.” While NBC’s internal motives can’t be proven, the observable outcome (a notably aggressive, prosecutorial town hall that reduced some of the backlash toward the network) matches that prediction.

Because the central, testable part of the prediction—NBC would “come out swinging” in a highly aggressive, challenging format toward Trump during that specific town hall—was borne out by both supportive and critical contemporaneous accounts, it is reasonable to judge the prediction as right, even though the subjective motive (“penance”) cannot be definitively verified.

What actually happened in the Nov. 3, 2020 U.S. presidential election

• Joe Biden defeated incumbent President Donald Trump with 306 electoral votes to 232.
• Biden won the popular vote with about 51.3% to Trump’s 46.8%, a margin of roughly 7 million votes.

These figures are documented in official tallies and widely cited summaries of the 2020 election results. (cfr.org)

Was Biden’s win a “landslide”?

Multiple nonpartisan fact‑checking and analysis outlets explicitly concluded that Biden’s victory, while clear and decisive, does not qualify as a landslide by common historical standards:

• PolitiFact: Biden’s win “wasn’t a landslide, but it was decisive,” noting that 306 electoral votes and his popular‑vote margin are solid but not historically large. (politifact.com)
• FactCheck.org similarly found that neither Trump’s 2016 win nor Biden’s 2020 win at the 306–232 level counts as a landslide by customary measures, placing such results well below classic landslides in U.S. history. (factcheck.org)
• Political science and historical discussions typically reserve the term landslide for elections with overwhelming electoral and/or popular‑vote margins (e.g., Johnson 1964, Nixon 1972, Reagan 1984), far exceeding Biden’s 2020 margin. (presidency.ucsb.edu)

Evaluation of the prediction

Jason’s prediction had two key components:

1. Trump will lose to Biden.

   • This was correct: Trump did lose, and Biden won both the Electoral College and the popular vote. (cfr.org)

2. The result will be a landslide.

   • By widely used historical and analytical standards, Biden’s 2020 margin is decisive but not a landslide; expert analyses explicitly reject calling it a landslide. (politifact.com)

Because the prediction specifically claimed a landslide (not just a clear win), and that core qualitative part did not occur by prevailing definitions and expert assessments, the overall prediction is best scored as wrong rather than fully correct, even though it correctly anticipated the winner.

Joe Biden did win the Nov. 3, 2020 U.S. presidential election, but the size of his victory was not "enormous" by historical standards.

Key facts:

• Biden won 51.3% of the popular vote to Donald Trump’s 46.8%, a margin of about 4.5 percentage points and roughly 7.1 million votes. (en.wikipedia.org)
• He won the Electoral College 306–232. This Electoral College margin is relatively modest historically and comparable to Trump’s 2016 result, which fact‑checkers have explicitly said was not a "massive landslide." (en.wikipedia.org)
• Several decisive swing states were very close: Arizona, Georgia, and Wisconsin were all decided by under 1 percentage point, with combined margins in the tens of thousands of votes. (en.wikipedia.org)
• Election analysts generally characterize Biden’s win as clear/decisive but not a landslide; it falls near the middle of post‑1960 presidential victory margins, not at the “enormous” end. (politifact.com)

Thus, while the prediction correctly anticipated that Biden would win, it specifically forecast an “enormous” victory—implying a landslide‑scale margin—which did not occur. The winner was right, but the predicted magnitude of the win was substantially overstated, so overall the prediction is best judged wrong.

Why this prediction is rated “wrong”

1. No Biden “landslide” in the actual 2020 result

   • Final outcome: Joe Biden won 306 electoral votes to Donald Trump’s 232, the exact reverse of Trump’s 2016 total. (en.wikipedia.org)
   • Popular vote: Biden 51.3% vs. Trump 46.8%, a margin of about 4.5 percentage points. (en.wikipedia.org)
   • Political science and media usage of “landslide” generally implies an overwhelming or unusually large margin; there is no fixed threshold, but it connotes a decisive blowout well beyond a typical competitive race. (en.wikipedia.org)
   • A 4.5‑point popular‑vote margin and a 306–232 Electoral College result (with several tipping‑point states decided by less than 2 points) are widely regarded as a solid but not overwhelming win. This falls short of what most observers would call a “landslide.”

2. Trump did significantly cut Biden’s polling lead after mid‑October
   The prediction added that Trump was “running out of time to change the polls,” implying the October polling picture would mostly hold.

   • Around three weeks before Election Day (mid‑October 2020), national polling averages had Biden ahead by about 10 points (e.g., RealClearPolitics and FiveThirtyEight both had Biden +10.6 on Oct. 13). (reddit.com)
   • In the final days before the election, a composite of major polling averages showed Biden ahead by roughly 7.9 points nationally (51.4% to 43.5%). (en.wikipedia.org)
   • The actual popular‑vote margin was Biden +4.5. (en.wikipedia.org)
   • From mid‑October to the vote count, Trump effectively narrowed Biden’s advantage by about 5–6 points relative to early/mid‑October polling, which is a substantial shift in the trajectory, even though it wasn’t enough to win.

Conclusion
The prediction contained two key claims:

• The race was “tracking toward a Biden landslide.”
• Trump had too little time left to significantly change the polls.

Biden’s eventual victory was neither a landslide by common standards nor a freezing of the mid‑October polling picture; Trump did meaningfully erode Biden’s lead before Election Day, though he still lost. Therefore, taken as stated, the prediction is wrong.

Donald Trump did not resign during the lame-duck period after losing the 2020 election, nor did Mike Pence become president and issue Trump a pardon.

Key facts:

• Joe Biden won the 2020 U.S. presidential election, and Donald Trump served out his full term until January 20, 2021, when Biden was inaugurated and Trump’s presidency ended normally; Trump never resigned from office during the transition period.
• Mike Pence never assumed the presidency and therefore never had the opportunity to issue a presidential pardon to Trump.
• No record exists of Trump receiving any presidential pardon (from himself, Pence, or anyone else) for federal crimes in connection with his term.

Because the prediction required both (1) resignation during the lame-duck period and (2) a subsequent Pence pardon for Trump, and neither occurred, the prediction is wrong.

Donald Trump did lose the 2020 election to Joe Biden, so the condition for the prediction was met. However, after that loss:

1. He did not resign before the end of his term. Trump served out his first term through January 20, 2021, delivering a standard farewell address on January 19 and leaving office when Biden was inaugurated, with no resignation or handoff of power to Mike Pence for purposes of a pardon deal. Sources: coverage of Trump’s January 19, 2021 farewell address and timelines of the end of his first administration. (en.wikipedia.org)

2. He did not secure broad preemptive pardons for himself or his family. Despite extensive speculation that he might issue self‑pardons or blanket pardons for his adult children and son‑in‑law, his final clemency list on January 20, 2021—143 pardons and commutations—did not include Trump himself, any of his children, Jared Kushner, or Rudy Giuliani. Multiple contemporaneous reports explicitly note that he “didn’t pardon his family” and “did not issue a preemptive pardon for himself,” and published lists of recipients likewise show no Trump family members. Sources: reporting on Trump’s final-day clemency (Forbes, Guardian, Snopes, Newsweek) and official recipient lists. (forbes.com)

Because Trump neither resigned nor arranged sweeping self/family pardons before leaving office, the specific scenario Jason rated as ~50% likely did not occur. The prediction therefore counts as wrong in outcome terms, even though it was framed probabilistically.