How to Fix Government, Twitter, Science, and the FDA, Four Years Later 如何修好政府, Twitter, 科学和 FDA, 四年之后

Modern nation-states cannot be forked. You can start a company. You can start a currency (Bitcoin proved that in 2009). You cannot peacefully start a new country. Governance has no "doc.new" button, so all political energy collapses into 51 / 49 reform fights inside existing institutions.

• i51% democracy is 49% dictatorship. The losing 49% has policies imposed on them they did not vote for, and the winner's campaign promises are not legally binding.
• iiInstitutions are run by inheritors, not founders. Founders combine legitimacy and competence; inheritors keep legitimacy and lose competence. Look around 2024-2026 — Boeing, Intel, the FAA after the Newark meltdowns, the Secret Service after the Butler attempt — all "maintainable, not rebuildable."
• iiiThe school of fish punishes individual dissent. Going with the crowd, no one is accountable ("who could have known"); deviating alone, you are a deviation to be punished.

Exit beats voice. Starting a new country should be as low-friction as starting a company. V1 hunter-gatherer → V2 farmer-soldier → V3 digital nomad / sovereign collective. The internet multiplied every individual's microeconomic leverage by orders of magnitude; "voting with your feet" became nearly free.

• i"Nation-state" barriers are lower than they look. Most UN members have populations under 10 million, smaller than large social networks.
• ii"Forking a country" has happened in history. Israel (Herzl 1897), India (non-violent independence), Singapore (Lee Kuan Yew as founding CEO), the US (founders mostly in their twenties), all forks of the UK codebase.
• iiiSovereigns are already open for business. El Salvador adopted BTC as legal tender, Wyoming recognizes on-chain LLCs, Nevada cut a deal with Tesla on the Gigafactory, Miami's mayor took his salary in BTC.

The concrete recipe · Cloud first, land last. 1000 people aligned online → in-person meetup → co-living → crowdfund property → networked archipelago → diplomatic recognition.

Supporting framework · Cloud regulators beat land regulators. 100 percent democracy replaces 51 percent (opt-in is real consent).

Net effect · Convert governance from an inherited / coercive / geographic activity into a constructive / elective / protocol-based one.

Balaji called the principle right but missed the vehicle from both ends. At the moment he was taping (2022-10-20) he did not know that Musk would, one week later, complete the Twitter acquisition — and that single event reshaped every exit that followed; without it the meta-thesis cashes out far less cleanly. What he expected to drive change was a "highly aligned online community"; what actually drove it was the decision of individual charismatic operators (Musk, Bukele, Trump, Balaji himself). Exit in 2026 is oligarch-driven, not community-first.

The mirror miss runs the other way: sovereigns really did start adopting BTC — but existing sovereigns absorbed it, not new ones spawning out. US Strategic Bitcoin Reserve EO (March 6 2025); Texas SB 21 (June 22 2025, first state to fund with state dollars); Arizona and New Hampshire reserve legislation. El Salvador, under IMF $1.4B EFF pressure, downgraded BTC from legal-tender mandate to voluntary (Jan 2025). Bhutan sold ~70% of its stockpile to fund Gelephu SAR. UAE drew ~9,800 net millionaires in 2025 (#1 globally). The third pole didn't replace nation-states; nation-states ate it.

The FDA is unfireable, unaccountable to voters, opaque to journalists, and the primary customer of every pharmaceutical company. The industry self-censors because antagonizing FDA kills the pipeline. Regulatory harmonization gives the FDA a de facto global monopoly: most foreign regulators rubber-stamp its decisions.

• iThe 1920s used to ship. Banting and Best went from idea to Nobel to nationwide Eli Lilly production in three years under willing-buyer / willing-seller. Today that timeline is 10-15 years.
• iiFive-party payment hides every signal. Employer-paid insurance + AMA-CPT-RVU pricing + insurer + provider + patient. No one knows the price until after the procedure.
• iiiCOVID exposed the failure mode. FDA blocked tests, refused challenge trials, and "the science" flipped on masks twice without acknowledgment.

Use the exits that already exist (right-to-try, CLIA labs, LDTs, compounding pharmacies, off-label prescription, foreign jurisdictions) and build a cloud-regulator alternative on top: a real Yelp for drugs, decentralized peer review, longevity DAOs. Ambulatory care should be cash; insurance should cover the ambulance, not the appointment.

• iExits already work, quietly. Right-to-try, CLIA labs, off-label prescription save lives at the margin every day. They're the proof of concept.
• iiIndia already runs the parallel market. India's generics industry (Cipla, Sun Pharma, Dr. Reddy's, etc.) ignores American IP on generics; cash payment culture; Aarogya Setu national telemedicine. A billion patients with different incentives.
• iiiContinuous diagnostics ran ahead of regulators. Larry Smarr's integrome story, Mike Snyder's quantified self, then Apple Watch + Oura + CGM all bypassed the 15-minute clinic visit.

Existing exits · Right-to-try, CLIA / LDT, compounding pharmacies, off-label prescription, telemedicine from foreign jurisdictions.

New stack · Cloud-regulator drug reviews (real Yelp for drugs), longevity DAOs (VitaDAO, Molecule), continuous biomarker streams (CGM, multi-analyte panels, eventually full Snyder-style integrome).

Payment · Singapore-style mandatory HSA, catastrophic-only insurance, transparent cash prices for everything ambulatory.

Balaji saw the institutional capture of FDA correctly and proposed founder-mode reform. The 2024-2026 reality is that RFK Jr's HHS seized the "reform FDA" mandate but pointed it at anti-mRNA / anti-vaccine targets, not at compressing trial timelines — and burned through its own appointees doing it (Marty Makary resigned May 12 2026 after 13 months as Commissioner; Vinay Prasad exited CBER July 2025, returned two weeks later, exited again April 30 2026). The political slot Balaji wanted — more trials, less gatekeeping — got filled by populist anti-science. The FDA didn't get reformed; it became the prize in a culture-war takeover.

Where exit-as-medicine did cash out, it cashed out in two corporate forms — consumer-driven self-tracking (Stelo, Lingo, Blueprint, Function Health) and AI-driven big-pharma capture (Isomorphic Labs alone raised $2.1B in April 2025; Generate Biomedicines dosed Phase 3 in Jan 2026) — not as the DAO-governed alternative the recipe specified. Longevity DAOs underperformed catastrophically: BIO Protocol -98% from its January 2025 launch; VitaDAO has deployed ~$4.2M across 24 projects. The decentralized-science substrate underperformed centralized-AI by roughly 500×.

Every centralized truth-keeper (peer-reviewed journals, NIH, Wikipedia) has the same architecture: a small permanent class certifies what counts as a "reliable source", and politically charged topics rot first. The QAnon alternative is to trust nothing, which is just noise.

• iCapital-S vs lowercase-s science. Maxwell's equations are science. Last week's paper in Science or Nature is "science" with quotes. Real science is independent replication, not prestigious citation. Most papers are not reproducible; the data is often not even public; prestige substitutes for verification.
• ii"Public Health was political health." COVID exposed the failure mode: masks don't work, then they do, with no acknowledgment of the reversal. FDA blocked tests. The institution above accountability stopped being science and became propaganda. Fauci as the face of science was the meme that finally added some accountability.
• iiiWikipedia is the same pathology, smaller scale. A few hundred senior editors classify sources via the Reliable Sources policy. Mother Jones green, Coindesk red, social media universally disallowed. Technical pages stay clean; political pages rot. Same gatekeeper-by-prestige logic as journal peer review.

Crypto-science. Move from prestige to replication, from gatekeeping to verification. Papers, code, data on-chain. Citations become import statements; replication becomes a download; one paper composes with another like DeFi protocols compose.

Ledger of record. Facts, identity, instrument hashes signed and timestamped on-chain.

Satoshi-anon. Independently verifiable signed assertions, not blind skepticism.

Wikipedia fork. Explicit editor identity (Google Knol style); articles become unit-tested computable artifacts.

• iReproducibility crisis is well-documented. Ioannidis's "Why Most Published Research Findings Are False" became foundational; the Reproducibility Project: Psychology found roughly a third of effects replicated. The disease is acknowledged.
• iiPreprint culture already arrived. bioRxiv, medRxiv, arXiv normalized open posting before peer review. Gatekeeping was already weakening before crypto.
• iiiBitcoin holds hundreds of billions in consensus across rival tribes without weapons. As a proof that decentralized truth-keeping can work in adversarial conditions, the machinery is there.

DeSci layer · VitaDAO, ResearchHub, Molecule, d.com style decentralized science protocols. Funding, IP, peer review, and data publishing on-chain. Composable papers: every result publishes its code and data with signed hashes; downstream work imports rather than re-derives. Instruments stream signed hashes on-chain for chain-of-custody. Wikipedia fork with explicit editor identity (Google Knol style), unit-tested computable articles.

The 2022 interview happened about six weeks before ChatGPT shipped, and the entire LLM era is missing from his frame. The actual fix to centralized truth-keeping came from more centralization, not less: a handful of California labs trained models that now mediate every lookup, write half the code on Earth, and accelerate biology directly.

Concretely, five closed labs out-shipped open communities on every scientific milestone that mattered: AlphaFold 3 (May 2024), AlphaProteo (Sept 2024, 3-300× binder affinity), GNoME (2.2M new stable crystals), AlphaProof + AlphaGeometry 2 (IMO 2024 silver), GraphCast / GenCast (weather), RFdiffusion 3 (Dec 2025). Isomorphic Labs alone closed Lilly ($1.7B) + Novartis ($1.2B) deals in Jan 2024 and raised $2.1B in April 2025 — second-largest biotech round ever, 17 active programs by Feb 2026. Pharma capital flowed to closed AI in billions; to DeSci in low single-digit millions (VitaDAO deployed ~$4.2M across ~24 projects, treasury ~$55.8M).

The grim joke: the product closest to Balaji's "composable papers with code" — Papers With Code — was sunsetted by Meta on July 24 2025 without notice, taking 9,327 leaderboards and 79,817 paper-code links offline. The real "Wikipedia fork" turned out to be LLMs themselves: Grokipedia (xAI Oct 2025) is a footnote (~35K daily visits after the launch spike), while ChatGPT / Claude / Gemini answer most of what people used to look up on Wikipedia.

The direction that went most "Balaji-shaped" is math and proofs: Lean as a machine-checkable substrate, AI agents formalizing proofs at scale (Terence Tao publicly using Claude Code with Lean in late 2025). But the agents are Claude and o-series; the verifier is Microsoft-funded Lean; zero of it is on-chain or token-incentivized. The shape of the answer matches; the topology is the opposite. He bet truth-keeping would decentralize via crypto; in 2026 it concentrated via AI.

The 2021 deplatforming of Trump proved that a small number of US tech CEOs can globally silence anyone, including a sitting US president. Once that seal is broken, every smaller head of state is on notice.

• i"Real name" is state name. The name on your ID is a surveillance tag built for the paper world. Gives away too much (stalker-friendly), gives back too little (no programmatic interface).
• iiCentralized messaging can be MITM-attacked at the server. Even "end-to-end encrypted" services trust the server in the middle; only locally-held private keys give verifiable end-to-end.
• iiiGen Z already solved part of this informally. Finsta (fake-name "real self") vs Rinsta (real-name curated) shows search-resistant identity as natural adaptation.

Pseudonymous economy. ENS / Farcaster / cryptographic handles replace state names; on-chain reputation accumulates across decades. Decentralized social (in 2022 Balaji specifically named Farcaster; Lens, Nostr, Bluesky emerged after) routes around the choke point. The cryptophone (Solana Saga) becomes the sovereign device.

• iPseudonymity already sustains real economies. Satoshi shipped Bitcoin under a pseudonym; the network runs at hundreds of billions of dollars without ever proving who he was.
• iiCryptographic identity is portable across decades. A long-running ENS or PGP key accumulates reputation no employer or state can revoke.
• iiiSelf-custody is the only sovereign primitive. If you don’t hold the keys, you don’t own anything; this generalizes from money to identity to messages.

Identity · ENS handles, multiple personas, search-resistant defaults. Comms · cryptographic E2E messaging keyed to local private keys. Social · Farcaster (in 2022 the only one Balaji named; Lens / Nostr / Bluesky followed), tokenize reputation, portable graph. Hardware · Solana Saga as the sovereign device.

Speech freedom in 2026 is oligarch-driven, not protocol-driven. The vehicle for exit was a person — Musk — buying the centralized platform, not a thousand users migrating to a sovereign one. Balaji's framework assumed coordination from below; the actual fix came from concentration at the top. The decentralized social stack was, in the end, optional.

The deeper miss runs through the same diagnosis: Balaji's 2022 threat model was platform CEOs, but the 2024-2026 threat is legislatures and courts — protocols don't route around statutes. EU Digital Services Act (€120M fine on X Dec 2025; preliminary finding April 29 2026 Meta breached DSA on under-13s, potential $12B); UK Online Safety Act (age-assurance live July 25 2025, Ofcom opened 90+ investigations by Feb 2026); SCOTUS upholding Texas age-verification 6-3 in Free Speech Coalition v. Paxton (June 27 2025), with 21+ states now able to enforce; Australia's under-16 social media ban in force since Dec 10 2025, requiring facial-estimation or ID upload across every major platform. The structural force in 2026 is mandatory ID at the network edge — the inverse of the pseudonymous economy Balaji predicted. The protocol-pseudonym stack didn't lose to Big Tech; it lost to age-verification statutes. The TikTok USDS JV (Jan 22 2026, Oracle / Silver Lake / MGX taking 45% combined, ByteDance kept at 19.9%) shows the same pattern on a different axis: the foreign-app problem was solved by forced incorporation, not by users migrating to a decentralized alternative.