Vitalik Buterin Walked Away From Billions of Dog Coins — Then Mapped Ethereum's Next Decade

Lex Fridman opens with the strangest fact in crypto that week. In August 2020 an anonymous founder known as Ryoshi launched Shiba Inu as a Dogecoin knockoff. By 10 May 2021 it had a market cap above thirteen billion dollars, and in a stunt the project sent Buterin half of the total supply. Fridman recites the headline version: Buterin burned 90 percent of it, then worth about $6.7 billion, and donated 10 percent, about $1.2 billion, to an India COVID-19 relief fund, saying he did not want to be the locus of that much power.

The backstory, Buterin says, starts with actual dogs. Dogecoin was a 2014 joke by Jackson Palmer that grew a community. Around 2016 Buterin put about $25,000 into it, wondering how he would explain "dog coins" to his mother; it became one of his best investments. After Elon Musk started talking about Doge in late 2020, the market cap shot toward $50 billion. Stuck in Singapore during COVID, Buterin watched the price jump from about 0.8 cents to seven cents in a day, sold half for $4.3 million, and donated the proceeds to GiveDirectly — just before it fell back to four cents. Then it ran again. People who had never heard of Ethereum had heard of Doge. The Shiba crowd's logic, as he heard it, was that if the leading dog token was worth $50 billion, the second deserved at least several billion — so they gave him half the supply.

They were not the first. A Chainlink competitor called Teller had dumped about $50,000 of tokens into his hot wallet and sent a Twitter army to declare that Vitalik was one of them. He sold the tokens on Uniswap in public. The Shiba team was cleverer: they sent the coins to his cold wallet. In crypto, he explains, an 80-digit private key is what actually owns the money. A hot wallet keeps that key in memory on a phone or laptop. A cold wallet keeps it on paper, or on a machine that never touches the internet. His setup was a laptop sitting in Canada plus two pieces of paper, one on him and one at home; add the two numbers and you get the key. COVID travel rules made that wallet almost unreachable. He is not sure the Shiba team knew that. They may just have been lucky.

Leaving billions unused started to look like a waste, so he powered through. He called family to read the Canadian number, bought a fresh laptop at Target, added the two secrets, and spent two days moving about 325,000 ether into a new multisig — the sort of procedure, he says, that ten years earlier would have lived only in a cyberpunk novel. He wrote his own JavaScript confirmations because Gnosis Safe and the Status wallet did not play well together, audited the new wallet, and called the whole thing a marathon. On paper the dog tokens were worth about seven billion dollars; in a thin market he had no idea whether selling a million, or ten million, would crash the price, or whether $200 million was even gettable. He later clarifies that he donated about 20 percent and sold 80 percent: one batch to the India COVID group, another to an organization he would not name because they wanted to announce themselves. A few weeks on, he does not really regret it. Traders at the charities extracted more than he expected, so he could have coordinated better — but he also finished most of the giving a few days before the crash. Parallel universes where he hesitated look worse.

Asked what principle survived the Shiba episode, Buterin reaches back to 2014. In Bitcoin Magazine he published a piece whose title he remembers as something like "Markets, Institutions and Currencies: A New Form of Social Incentivization." The idea was that you could have coins that stand for causes, and people would buy and accept them because they support those causes. Dog-coin philanthropy made a cartoon version of that essay real. Crypto, he says, is not only a slightly better savings technology. It is a chance to build digital institutions that serve the public good in new ways.

Public goods matter even more on the internet. Every Lex Fridman episode sits on YouTube for anyone to watch; gating it would shrink the impact. The tension is that YouTube feels open and is still one company. Fridman says he first felt that YouTube was not forever when the Joe Rogan library left for Spotify — the same jolt as realizing fiat money is centralized. Buterin's philosophical target is to decouple "public goods," which he calls the lifeblood of modern civilization, from the assumption that one central organization can represent the public and impose its idea of the good. Talk of public goods often arrives with baggage of centralization or conformism. Often the most important ones, he says, are made by the crazy individual who disagrees with everyone else.

The synthesis he wants is decentralization and open source without naivete. These things still need a sustainable way to pay the people who make them, and that funding channel must not become a vector of control. In Ethereum that experiment includes Gitcoin Grants quadratic funding. Uniswap's DAO sits on a huge treasury whose future uses are still unknown and could be "really good and amazing." He does see Ethereum as a way to fight for some social causes — an open financial system anyone on earth can join, five people in five countries able to organize a project. The other half of the magic is that it remains an open platform. What lives on Ethereum is whatever the community makes of it.

Fridman asks for best and worst cases of government regulation. Best case, Buterin says, is that blockchains keep prospering, scalability gets solved, and the applications people have talked about for years actually run on-chain instead of being priced out by five-to-ten-dollar fees. DAOs invent new ways to organize, artists find new funding, and enough of the public can see that crypto is doing good that it gets incorporated the way the internet was. There will always be people who want to surveil everyone. If the empowering effect is larger and visible, that is still a good world. Corporate tax as a revenue stream might even fade if companies are replaced by DAOs — unless DAOs encode some bias toward the global public good.

Worst case is a moral panic. Governments cannot, in his view, ban crypto hard enough to stop blockchains from existing. They can marginalize it: ban exchanges, cut every fiat on-ramp, forbid mainstream employers from paying or accepting crypto, and turn the whole thing into a niche counterculture with far less impact. Fridman notes the other tension — billionaires and firms such as Tesla embracing Bitcoin while governments resist. Respectable money, he suggests, is a signal that the revolution is staying. Buterin is hoping for the good scenario.

On the old charge that crypto funds crime, he does not deny the risk. He thinks the rest of technology is moving the other way: in-person surveillance rises every year, and meatspace crime gets harder to get away with. A world so dark that criminals act with impunity seems unlikely. A world with no privacy, and no ability to act outside mainstream institutions — including corporations, not just governments — is more realistic and scarier. Many governments, he adds, are already anxious about sovereignty. The Indian government is on team India, the Russian government on team Russia; they do not want a U.S. backdoor into everything through foreign tech firms.

Payment processors, he notes, restrict a much larger set of activities than the law does. They are conservative because every extra layer is afraid of the layer below, and they also carry their own morals. They make life hard for sex workers and for psychedelics research, including work that is entirely legal. Fridman had been talking to people in that gray zone; Buterin's punchline is institutional, not romantic. "Reducing the number of intermediaries is definitely normally a good thing."

They have been de-emphasizing the "Eth 2.0" brand, Buterin says, because the original picture was one grand cutover to a new chain. The roadmap became incremental: proof of stake over time, sharding over time, a user experience closer to the hard forks Ethereum already ships. The two flagship features remain. Proof of stake is a consensus *mechanism*, not just an algorithm — it includes people and incentives. Nodes still have to agree which blocks arrived, in what order, and that accepted blocks do not get reverted. Bitcoin, today's Ethereum, and most chains still use proof of work.

Both proofs exist to buy what he calls economic Sybil resistance. You cannot do one-person-one-vote on an open network; an attacker can spin up ten billion virtual nodes. So influence is proportional to economic resources. Proof of work proves those resources by burning specialized hardware and electricity around the clock. ASICs take a huge cut, warehouses fill with machines, and about half the cost is hardware, not power. He had just seen a story about Iran wanting to shut Bitcoin mining because it was outbidding nearby towns for electricity; Chia's hard-disk mining had created its own shortage. Bitcoin and Ethereum each issue around 4 percent of supply a year to miners — Ethereum about 4.7 million ether on a supply of about 115 million. Proof of stake should need only 500,000 to one million new ether a year, because validators can run on ordinary computers.

Security, he argues, is not weaker. About five million ETH was already staking, on the order of $15 billion — more, he believes, than the cost of attacking Bitcoin. Many attacks can be answered with automatic slashing: if validators sign conflicting messages to revert a finalized block, the protocol burns *their* coins. A censorship attack is messier. The censored minority would start another chain, and the community would have to choose it in a soft fork, after which the attackers lose coins on the new chain too. Whether that social coordination would actually work is untested. Communication happens on Twitter, Reddit, GitHub, Ethereum Magicians, ethresear.ch, Telegram, and Signal. A serious attack, he thinks, would be visible in a way that political facts are not. Recovering in proof of work is worse: you cannot identify which miners attacked, so you either keep all of them or change the hash algorithm and eject all of them.

Some people call miner extractable value an existential risk to Ethereum. Buterin says the problem is the same in proof of work and proof of stake — call it block-proposer extractable value if you want. Whoever chooses which transactions enter a block, and in what order, can earn more than fees. On Uniswap, if ETH/USDC was 2700 last block and the market is now 2680, a proposer can eat the automated book between those prices, buy back on the open market, and keep the spread times the depth. Front-running is ruder: if someone sends a buy of five ETH with a $15,000 cap, you buy first and resell at $14,999.

That degrades user experience, but the deeper fear is economies of scale. Collecting fees is a simple sort: take the highest-paying transactions. Extracting MEV wants sophisticated algorithms. Wealth buys talent, talent buys better bots, and the rest of the capital buys hashpower or stake so those bots run more often. Mining or validating could centralize. The ecosystem's best reply, and the path Flashbots is already on, is: if you cannot eliminate the centralization, firewall it. Build a market where "searchers" do the hard work of packing profitable bundles, and block proposers stay dumb nodes that ask every searcher what they will pay and take the highest bid.

It is an economic-design motto he likes. Sometimes centralization will not go away. You can still contain it, direct it, and keep it off the parts of consensus that actually need to be decentralized. He does not call MEV existential. It is a risk, possibly a severe one, and Flashbots could have a fatal flaw. The working attitude is that it is a problem, work is being done, and so far it is being solved.

Layer-1 scaling, Buterin says, means making the chain itself process more without each node becoming a supercomputer. The dumb version is bigger blocks. He has been called a small-blocker and a big-blocker; he says he has been a medium blocker all along. The trade-off is between making it easy to *write* the chain (send transactions) and easy to *read* it (run a node and verify). If reading is too expensive, everyone trusts a few readers, who can change the rules. If writing is too expensive, everyone flees to centralized second layers, which also kills self-sovereignty. Bitcoin originally had no real block-size limit, only an accidental 32 MB cap from peer-to-peer messages. In 2010 Satoshi cut it to 1 MB as a safety measure. Most people, Buterin thought, treated that as temporary. As usage climbed from 100 to 250 to 500 KB per block, a faction decided the cap should never rise.

Compromise attempts piled up — a 20 MB proposal, then "2-4-8," which began as a small-block negotiating position and was rejected once big-blockers agreed. Small-blockers valued 1 MB for two reasons: they really believe in cheap verification, and many believe in a norm of never hard-forking. A soft fork keeps old software valid on the new chain; a hard fork requires an upgrade. They call soft forks less coercive. Buterin disagrees completely. Soft forks, he says, force everyone who disagrees to go along by default. Hard forks force a choice: accept the change or not. He grants that soft forks are harder to abuse for something like an unexpected supply increase. Segregated Witness was the clever trick: signatures move outside the block, an extension block appears that old software does not even see, and capacity rises to about 2 MB while still measuring under 1 MB by the old rules. He favored the big-block side the whole way. Hard forks do raise the risk of a split. A split from a bug is horrible. A split from a political disagreement, he thinks, is better than one side being forced to swallow a majority it hates — and on a blockchain, walking away costs far less than splitting a country.

He is disappointed in what happened to Bitcoin Cash. Opposition parties, he says, often have legitimate grievances and still lack the capacity to govern; their culture is built for resistance, not trade-offs. Craig Wright walked into that vacuum. Fridman notes that Wright keeps claiming to be Satoshi and suing people. Buterin, on tape, addresses Wright's lawyers: he still thinks their client is a scammer, so sue him. He compares Wright to a Donald Trump figure — not very intellectual, good at saying what resentful people want to hear. After the block-size war, big-blockers felt Bitcoin's original peer-to-peer-cash vision had been replaced by an elite of Core developers waving an unproven Lightning Network. A lot of that anger, Buterin thinks, was justified. Wright repeated the talking points and added the Satoshi claim as marketing salience.

The tell, for Buterin, was technical. Wright mixed up 256 bits with 2^256 bits — the difference between eighty and an 80-digit number, between a tiny quantity and more information than exists in the universe — and built an argument that Bitcoin's elliptic curve was pairing-friendly on that confusion. Pressed, he equivocated. Buterin called him out, including standing up at a conference in Seoul to ask why a fraud was speaking. Some big-blockers were angry at him then; they later expelled Wright, who split off BSV when the community refused a change he wanted. Bitcoin Cash then split again over a plan to redirect 12.5 percent of miner revenue to developers. Buterin had wanted a principled coin that tried to be Bitcoin with consistent big-block values. Maybe, he adds, he should stop expecting projects he has no involvement in to care about his values.

If you scale a blockchain the dumb way, by turning up the parameters, you eventually get about twenty computers running everything. Sharding changes the architecture so each node stores and processes only a slice. The metaphor is BitTorrent: there is no full node that holds every movie, and a design that required one would have zero censorship resistance. Blockchains add a harder demand. They must agree on *exactly* which data was published, and that it is correct. The paradox: you want 10,000 transactions per second, each machine can personally check 100, and every machine still needs a guarantee about the other 9,900.

The tricks stack. Random sampling on a proof-of-stake chain with 10,000 validators assigns a random hundred to each block. They sign, the signatures combine, and everyone else verifies the aggregate rather than the block. If the whole set is mostly honest, random committees are mostly honest, and an attacker cannot stuff a single committee. ZK-SNARKs go further: a short proof that a computation was run correctly. Even a fully evil committee cannot forge a valid proof for a bad block, which is a step up from a 51-percent-honest assumption. Data-availability sampling checks that the bytes were actually published. Together, an individual can be convinced the whole distributed chain is correct without personally checking more than about 1 percent of it. The current proposal turns on 64 shards, with room in the data structure for 1,024; too many shards make the management logic itself expensive. Sixty-four can be hard-forked up later.

Maximum scale, he says, comes from combining sharding with rollups. Users send transactions to an aggregator — theoretically anyone can be one, so sending stays permissionless. The aggregator strips anything not needed to update "state" (balances, code, contract memory), then compresses what remains. His own 20-byte address can become index 1,874,224 in a tree, three bytes instead of twenty. On-chain data can drop by about a factor of ten. Computation leaves the chain two ways. A ZK-rollup posts a SNARK that the result is correct. An optimistic rollup lets someone assert a result; only if someone disagrees do you publish everything and replay on-chain, and the loser burns a lot of money, so disagreements are rare and expensive. Core principle: put a compressed blob on-chain so everyone can keep up with state, prove the computation off-chain, and keep about 10 percent of the data and 1 percent of the computation on-chain. Scalability rises about a hundredfold. Loopring, a payments ZK-rollup, already cut fees from about five dollars to less than five cents. Fully EVM-equivalent rollups, he expected, were months away. A hundred from rollups times a hundred from sharding is 10,000 — hundreds of thousands of transactions a second.

In the long run he is on record that ZK-rollups win. Once you trust the cryptography they are conceptually simpler: no challenge game, no week-long withdrawal window while someone might dispute a block. As soon as a ZK block lands with a proof, you know it is correct and can exit immediately. They are not winning everywhere yet because SNARKs are still new — theory a decade ago, excitement at Bitcoin conferences about eight years ago, first usable on Ethereum three or four years ago, efficient enough for serious work only in the last year and a half. Fridman plugs Buterin's essays at vitalik.ca. Then they turn to other Layer 2s.

Polygon is popular, and Buterin draws a hard line between rollups and sidechains. If your coins sit in Loopring, Optimism, Arbitrum, or zkSync, then even if everyone else in that ecosystem hates you, you can still force an exit onto Ethereum mainnet. It may be expensive. You can still do it. Polygon is a sidechain, secured in part by its own proof-of-stake and its own token. If 51 percent — he even says 70 percent — of Polygon holders wanted your coins, they could take them, and the supply, he thinks, is not that widely distributed. Why is it faster? In part because it is more willing to take centralization trade-offs; in part because a hundred copies of a sidechain would sum to more throughput even if each copy were only as scalable as Ethereum.

He still respects the work. Ethereum needs to scale *now*. Apps that cannot find an Ethereum-friendly option will not wait twelve months; they will go to Binance Smart Chain or something with no alignment at all. Optimism was not live. Polygon was. They chose functionality first and security later, and he has talked with them about becoming a rollup, or at least adding a rollup chain — a follow-through he thinks they should make. The broader engineering moral is one Ethereum tries to live by: when you ship the crappy solution, pick a crappy solution that is forward-compatible with becoming good. The first sharding version may well lack SNARKs and data-availability sampling, but the sockets for those parts should already be there. Skip that thinking and you back yourself into a corner, then spend years removing features you cannot sustain — he cites raising gas costs on operations that were too cheap relative to the CPU they actually burned.

The Merge is how proof of work goes away. The Beacon Chain launched in December 2020 and had been running nearly six months, coming to consensus only on itself, on purpose: time to mature, time to prove to the community that proof of stake is real. Earlier Merge designs asked users and clients to migrate to a new chain. The new design embeds the old chain inside the new one. From the transition block, every proof-of-stake block contains an "execution" block of the Ethereum people use today. Those execution blocks no longer need proof of work; they need to be nested in a proof-of-stake block. A chain inside a chain is slightly ugly architecture and very kind to users: clients keep following the same rules, except they check embedding instead of hashes. Pre-merge history will not live in the new chain. Nodes around the Merge will still sync the proof-of-work history, then stop bothering. Archives can sit on The Graph or BitTorrent. Smart contracts do not strictly need it.

Why so slow? Technical complexity, he says, more than social complexity. They have been lucky: little drama. Proof of stake and sharding were on the roadmap almost from the start — the first proof-of-stake blog post is January 2014, two months after the project began. The DAO fork also sorted the tribes. People who could not stomach that fork went to Ethereum Classic; they tended to be more Bitcoin-ish, and Bitcoin-ish people tend to like proof of work. Classic is not switching. By Beacon Chain launch the remaining community was strongly for the switch. The timeline question every engineer hates: early 2022 was the realistic date, with an optimistic case still in 2021. A good sign would be a spec of both the transition and the post-transition world, a testnet that does both, heavy audits, then a run on an existing testnet such as Ropsten or Rinkeby before mainnet.

Testnets are where the scares happen. On Medalla in mid-August 2020, clients interpreted rules differently, one got kicked off, and recovery took weeks — partly because testnet coins are worthless, so nobody rushes to upgrade. Finalization failed again in October. Developers learned how to optimize, and they watched the full life cycle of more than a third of validators going offline: finalization stops, inactive stake is "weaked" (burned without using the word slash) until participants are two-thirds again, then the chain resumes. Seeing those edge cases live, Buterin says, made people more comfortable about mainnet.

On 24 April 2021 a bug in Prysm stopped roughly 70 percent of Beacon Chain validators from producing blocks. Attestations — votes that actually finalize — still worked, so the chain stayed quite stable. Nimbus, Teku, and Lighthouse kept proposing. That, for him, is the lesson. Ethereum does not have Bitcoin's client monoculture. Multiple implementations speak the same protocol. When they disagree because of a bug, it is a consensus failure, and those are serious. Bitcoin's monoculture makes them rarer, though not absent: Bitcoin had a consensus failure between two versions of the same client in 2013. Multi-client architecture, Buterin thinks, has saved Ethereum more than it has hurt it.

The 2016 "Shanghai DoS attacks" — named because they started on the first day of the conference in Shanghai — produced blocks that were deadly slow for one client and fine for the other. The two clients then were Geth and Parity (the top three by 2021, he says, were Geth, Nethermind, and Besu). The attacker could not brick both at once. Miners flipped between implementations, the chain survived a month of hammering, and the teams were forced into a sprint that burned out the DoS bugs. Bitcoin's 2010 overflow bug is the counterexample: a transaction with two outputs of about 2^63 satoshis wrapped past 2^64, nodes thought the inputs covered it, and billions of bitcoin appeared from nowhere. It took about twelve hours to notice and fix. In a multi-implementation world, one client might have bugged and another would have rejected the wraparound, creating a messy consensus failure that at least sounded the alarm and gave social permission to pick the correct chain.

The model still draws fire. Bitcoin people talk about consensus-failure risk. VC types call three funded teams expensive and "unfocused" — they love the word focused. "Ethereum is not that," Buterin says, "but it's amazing despite not being that." A year of running the Beacon Chain as a trial before all of Ethereum depends on it is, in his view, time well spent: you learn what staking actually feels like, and what the incentives actually are.

Why does the world need Ethereum if Bitcoin plus Lightning "fixes everything"? What always attracted him to Bitcoin, Buterin says, was decentralization: open systems anyone can join, that do not die when the founder gets bored, that resist operators breaking the rules. Those principles apply to much more than money. Bitcoin is a blockchain for money. Ethereum was a general-purpose blockchain from the start. Ether is an asset, and you can also build DeFi, ENS, prediction markets, or non-financial records of whether a certificate was signed or a key revoked. Without a blockchain, what a decentralized computer network can do is limited; with one, many of those limits fall. Smart contracts are the core of that generality: base-layer logic that runs in the open, or at least produces zero-knowledge proofs that the rules were followed, so you are not constantly trusting a central actor.

He does not think money is the world's only problem, and he is not in the camp that replacing fiat with crypto would end wars. Seigniorage is maybe 5 to 10 percent of government revenue. Take that away from an "evil" state: does it stop bombing Afghanistan, or cut welfare? If you think the first, you have a very optimistic view of government. A decentralized open currency still has real value. Many places have money far less stable than the dollar. Bitcoin offers those people... Bitcoin, which is also volatile. Ethereum offers ether *and* stablecoins. Ideological purists will say dollars collapse too; the Venezuelan bolivar is still much worse. A global open financial system can help, but only if it is more than a currency, and only if fees are not five dollars a transaction — nobody will pay that for non-financial apps.

Elon Musk had said Doge should 10x block time, 10x block size, and 100x lower fees, "then it wins hands down," and joked that Buterin feared the Doge. Buterin wants the record corrected: he does not. He visited the original Doge in Japan; she was still alive. Ethereum events at DEF CON take Doge. He holds a bunch. Ethereum is, in spirit, a little bit of a dog coin. The technical objection is the same as Bitcoin's block-size war: cranking parameters without new architecture makes the chain harder to validate and easier to capture. If Doge bridged to Ethereum, people could trade it thousands of times a second inside Loopring. If Doge copied ZK-rollup tech on its own chain, that would be great too. Cross-chain bridges are still in their infancy. Doge merge-mines Litecoin, which looks worse in hindsight now that Doge is larger. Early Ethereum had BTC Relay, a contract that verified Bitcoin blocks, then fees got too high and interest faded. A SNARK-powered version could be cheaper, but moving assets needs *two-way* verification. Ethereum can verify anything; Bitcoin would need a (soft) fork. If Doge wanted a two-way fork, he would call it a lovely collaboration, and he thinks some multisig funds might even bounty it.

Do not, he says, hang a sophisticated or malevolent plot on Musk. He is a human who likes dogs, the way Buterin likes dogs. He launched a car into space for fun. More and more he lives by "the most entertaining outcome is the most likely," and Dogecoin is the most entertaining cryptocurrency at a moment when the industry is getting very serious. Fridman asks about Chainlink and oracles. Buterin thinks off-chain data is necessary — stablecoin prices, Etherisc-style crop insurance in Kenya or Sri Lanka, prediction markets. Chainlink occupies the fast, automated part of that design space; Augur occupies the slow end, where you can often just sell at 99 cents if reality has already resolved. He is glad the project exists and notes that its "frog army" on Twitter can get intense. He does not want oracles in Ethereum's base layer. If the chain itself asserted a USD/ETH price, it would be making geopolitical statements. Imagine a U.S. civil war with two dollars: whose dollar does the oracle pick? Keep Layer 1 a pure platform of deterministic rules, and leave oracles to Layer 2.

On Charles Hoskinson and the early Ethereum split, Buterin would rather let the books speak. People who follow Hoskinson closely have told him that 2021 Charles is different from 2014 Charles; 2021 Vitalik is different too. A 2030 reunion is possible without being best friends. Cardano, he says, really emphasizes big academic proofs; Ethereum is more willing to use heuristics because it is trying to do more, faster. IOHK research still produces interesting work. He personally thinks deep rigor is overrated, because protocols fail outside the model at least as often as inside it. Selfish mining in 2013 showed that Bitcoin's 50 percent honest-majority story drops toward 0 to 33 percent once you assume rational actors and a realistic network. Traditional consensus research barely modeled incentives; it assumed fifteen institutions you could sue offline. Crypto cannot sue. He did once find an inside-the-model bug in a proposed BitShares or EOS consensus around the end of 2017. Even then, what you are *trying* to achieve can never be fully written in formal language — the AI-safety people, he notes, keep rediscovering that a specification of what you want is insanely hard, and the more powerful the optimizer, the more careful you have to be.

He writes for curious outsiders, not for other people inside the academic system, which is why traditional academic types sometimes call him reckless. You probably want both kinds of analysis, in stages: quick-and-dirty public feedback, then more formalism. The internet has already changed collaboration; there is no reason the right structure for research is the one from fifteen years ago. It is healthy that ecosystems have different attitudes. Ethereum may be wrong about some things, and other chains can be the existence proof.

On AI safety his faraway impression is that AI may not be formal *enough*. Practice is "I found a couple of hacks, classification went from 0.684 to 0.773," often with no science for why. That is bad for legibility and for imagining failure. There is even a live debate over whether GPT-3-style models just keep getting smarter as you scale them, or whether some kinds of reasoning they will never acquire. Cryptography is a mix: very formal given its assumptions, very informal about whether factoring or elliptic-curve discrete log actually stay hard after decades of trying. At least the informal bets sit on a few building blocks many people watch, and ZK-SNARKs reduce a big statement to those blocks. Consensus remains more art than science because it is crypto-*economic*: it makes assumptions about people, and math will not prove whether 51 percent stay honest or whether the other 49 percent can coordinate a fork.

NFTs surprised him. On a tech-conference "overrated or underrated" panel he said overrated, and in retrospect he was quite wrong. He cannot personally relate to paying huge sums for uniqueness — in a mansion he would hang photocopies, or Nyan Cats, not a Mona Lisa. What makes him happy is the public-goods loop from the start of the conversation: if valuable work cannot pay its makers, it either does not get made or it gets made in a centralized substitute that concentrates control. Artists minting NFTs is another revenue stream for chronically underfunded creative work. As for "scam," he insists it is a gray area. BitConnect's 2017 three-minute-forty-eight-second rant at a Vietnam conference was the peak of pure scam coins, and there is less of that now even though crypto is bigger. Then comes a spectrum: technically fine projects with sketchy communities, nice communities attached to projects that cannot do what they claim, and actually good work. Bitcoin SV is his "80 percent scam" exhibit — a real chain with 512 MB blocks you could even use as a file backup, steered by Wright and a half-insane community making obviously false claims. Tron is a platform you can use that also plagiarized the IPFS white paper. He would not call Bitcoin or Litecoin scams; earnest people trying to make Litecoin be Litecoin is enough. A real scam is key people *intentionally lying*. Flawed or wrong ideas with good intent are not scams. Money just turns every tribal war into Mac versus PC with ten shares of Apple taped to every laptop — 1983 elites who spent $500 and now sit on $40 million, theorizing geopolitics. That is what energizes the armies.

Buterin has given to the SENS Foundation. His hope is that watching parents and grandparents die slowly disappears from public consciousness, the way getting lost in a city disappeared once everyone had a smartphone. He keeps a Nick Bostrom essay pinned on Twitter that treats death as almost unethical: there is a dragon murdering everyone, including your family, and not fighting it aggressively does not make sense. Over the last five years he thinks humanity has started to see the first cracks of victories. "Forever" needs a definition — supermassive black holes eventually fill the universe, so we have maybe 10^16 years to work. Getting to millennia will take an insane amount of work. After the first problems fall and people live to 150, ten slower killers will appear. Aubrey de Grey's longevity-escape-velocity argument is the comfort: those extra decades are time to fix the next crop. Do not underestimate ingenuity. ENIAC in 1950 to computers in 2020 is seventy years, and both he and Fridman likely have at least seventy left. COVID, he thinks, shoved radical biomedicine into public view and reminded people that bits and tweets do not save the world by themselves; vaccines did.

Late-twentieth-century ideology, as he reads it, said nature is good and disruption is bad. The right-wing version treats markets as nature; the conservative version treats traditional culture as nature. COVID flipped a lot of minds: nature is not safe, the default is tens of millions dead, and the way out is human ingenuity. That frame is friendlier to treating aging as an engineering problem. The default is that 7.8 billion people die, the last decade in debilitating pain, unless ingenuity stops it. Computational biology is turning the mess of the body into something you can point engineering tools at. The other track is less romantic about the meat vehicle: Neuralink-style interfaces, uploading, immortality in information space. He wants all the parallel bets — better bodies, better brain scanning, better cryonics, a last-resort freezer in which someone like Hal Finney might one day wake. Hard science fiction adds a practical reason to leave biology: Breakthrough Starshot-style probes to Alpha Centauri, four light-years away, perhaps by the 2060s, accelerated at something like 10,000 g to 20 percent of light speed. A human does not survive that. A robot might. Both men are in the camp that consciousness is a property of the algorithm, not of brain tissue; Fridman worries copies and infinite life would drain meaning. Buterin's historical parallel is philosophers who said war gives nations meaning — a view that ended with a "crazy Austrian" whose soldiers shot at Buterin's grandparents. Life stayed meaningful without existential combat. Business, ideology, and even Diablo hardcore mode can supply finiteness. He still talks to zero middle-school friends and two from high school. A person and their 25-years-later self, he recalls a study saying, are about as different as two people. You can have the sorrows that make meaning without anyone having to die.

On the meaning of life, money taught him a subtractive lesson. The most you get from it is fewer things to worry about, not more things to have. Non-zero savings mean you fear job loss less. Six months of runway makes it easier to leave a job that conflicts with your values. More money still and you can stop needing the work to be profitable, or pick lazier food and transport. Treat money as a machine that hands you stuff, and you are likelier to squander the opportunity. After money stops being the problem, meaning is. School, he notes, is totalitarian and also a post-scarcity utopia: twelve years with resources provided and choices made for you. Adulthood is the burden of freedom. Even his first five years of Ethereum were not that self-directed — speak in Korea, speak in Taiwan, finish this piece so the chain can launch, invent a proof-of-stake algorithm. COVID, holed up in Singapore, gave him alone time. He decided to make it count, discovered podcasts as a listener rather than a guest, and, on a recommendation from Karl Floersch at Optimism, burned through Hardcore History, about ten Fridman episodes, and the 45-hour *Rise and Fall of the Third Reich*.

The 1930s and 1940s fascinate him because they force the question of where evil comes from. The Holocaust happened; if you have coffee with people, a hundred times out of a hundred they seem nice. His own school years were kindness, which made later meanness harder to parse. Within six months of starting Ethereum there were fights over shares if it became a company; he suggested a nonprofit and somehow upset people. Watching crypto and reading that history, he arrived at a working intuition: most evil comes from fear, not greed. The DAO fork in 2016 was his personal low. After the hack and the controversial protocol change, Ethereum Classic split off. Ether started around $13; then one day ETH was $12.50 and ETC was $0.50, and people decided Classic was a Bitcoin-community psyop. Some of that was real; blaming "foreigners" is also what governments do to avoid facing dissent inside the tribe. People called for trademark lawsuits to stop exchanges listing Classic. Using the state as a weapon against another cryptocurrency, he says, was deeply unethical and against the freedom the project claimed to stand for. He played along to some extent and regrets the excesses. The Bitcoin block-size war had the same shape. When you are afraid that unless you act, your whole world collapses, it becomes easy to forget your principles. Fridman puts the moral as courage: in the face of fear, keep compassion rather than hate. Buterin agrees. Everyone has a capacity for evil and for love. The job is to build mechanisms and incentives that prefer the latter.