Part IThe Story
In March 2009, a 37-year-old Estonian programmer sat down in what he later called "a bland franchise eatery next to a noisy California freeway" to meet a young man whose blog he had been reading. He had flown a long way for the meeting. To be recognizable, the blogger wore a button. It read: Speak the truth even if your voice trembles.
The programmer was Jaan Tallinn. Six years earlier he had helped write Skype, the peer-to-peer telephone that let a student in Riga call her sister in Massachusetts for free and stay on the line for an hour talking about cafeteria fruit. Skype had been sold to eBay four years earlier for roughly $2.6 billion, and Tallinn was, by his own account, "in the process of scaling down my work" there and looking for whatever came next. The man with the button was Eliezer Yudkowsky, a self-taught AI researcher whose essays argued, at length and with a kind of patient alarm, that a machine smarter than its makers would not stay obedient by default, and that almost nobody was working on the problem.
"His name was Eliezer Yudkowsky," Tallinn wrote years later, "and we spent the next four hours discussing the message he had for the world—a message that had brought me to that eatery and would end up dominating my subsequent work."
Four hours. Traffic on the freeway outside. Two men at a laminate table arguing about the end of the human era.
Most of what is interesting about Jaan Tallinn follows from that afternoon, though not in a straight line. He left it convinced, and he remembers the conviction as a kind of embarrassment: "Why hadn't I figured this out? Why was nobody else talking about this kind of thing seriously? This was clearly a blind spot that I had fallen prey to." Over the next seventeen years he would co-found two research institutes in two Cambridges, sign every major open letter warning that artificial intelligence could kill everyone, sit on a United Nations advisory body, and spend his Skype money trying to slow down an industry he considered reckless. He would also become, almost against his will, one of the most successful AI investors alive. He was an early backer and board member of DeepMind. He led the Series A of Anthropic. In September 2026 the Wall Street Journal ran a headline about him that could serve as an epitaph for the whole category of person he had become: "He's Completely Terrified of AI—and He's About to Make Billions Off It."
When the physicist Steve Hsu, interviewing him in 2024, asked whether it was true that he was the world's greatest AI investor, Tallinn did not take the compliment. "Oh, I don't think so," he said. Then, after a beat, a correction that was also a confession: DeepMind and Anthropic, he allowed, were "going to play two notable entries in my portfolio."
The button is worth holding onto. A voice that trembles but speaks. A man who builds the thing and warns about the thing. A message, delivered in a franchise restaurant, that turned out to be load-bearing.
By the Numbers
Jaan Tallinn
$5,000Earned in 1989 from Kosmonaut, the first Estonian computer game sold abroad
$330Daily rate each Bluemoon founder took doing remote work for Tele2 as the company faced bankruptcy in 1999
$2.6BeBay's 2005 purchase price for Skype, when Tallinn sold his shares
~$200KInitial donation to co-found Cambridge's Centre for the Study of Existential Risk in 2012
$124MAnthropic Series A he led in 2021, at a reported $623M post-money valuation
€10.3BNet worth atop Äripäev's Estonian rich list, September 2026
The Computer Games Industry of Estonia
"I was born behind the Iron Curtain, in Soviet-occupied Estonia," Tallinn told the editor John Brockman in 2015, "and looked forward to a pretty bleak life in some scientific institute trying to figure out how to kill more Americans."
It is a joke, mostly. The timeline rescued him from it. "Luckily though, the Soviet Union collapsed shortly before I was ready for independent life." He entered the University of Tartu in 1990, the year the empire was coming apart and Estonia was clawing its way back to independence, and he studied physics there because he was already a programmer and "thought I should expand my horizons a little bit." He graduated in 1996 with a bachelor's in theoretical physics. His thesis considered how one might travel interstellar distances using warps in the fabric of spacetime. A young man in a country that had just recovered its own borders, writing about how to get past the ones the universe imposes.
But the programming came first, and it came out of scarcity. In 1986, at fourteen, Tallinn and a classmate named Priit Kasesalu started writing software for a small hardware company in Tallinn that assembled 8-bit computers, built around Intel 8080 processors, for use in Estonian public schools. The machines were slow. That, Tallinn would later argue, was the gift. "The nice thing about starting your computer career with computers that are really slow is that you have to do work in order to make them do something interesting." (Of his own children he observed, with the faint melancholy of a man who has watched abundance arrive: they "have a tough time starting programming because YouTube is just one click away.")
Then Tallinn met Ahti Heinla, who arrived already fluent in the Z80 chips that powered Yamaha MSX computers, and the three of them decided to write an action game for the IBM PC, a platform which at the time, as their old company website puts it with retro-futurist pride, "already boasted 'superb EGA graphics.'" A graphics man from Estonian State Television's computer graphics division—the website calls him "a hairy dude," and his name was Kaspar—offered to help with the art. The game was called Kosmonaut. In 1989 it was published in Sweden by Scandinavian PC Systems, the first Estonian computer game ever sold abroad. Tallinn was seventeen.
They were paid in Swedish kronor. "I think we made $5,000 based on that," he recalled, "which was an incredibly big sum back then." The ruble was in freefall. Hard currency, in that moment, was not merely money; it was a door. "Hence, we were hooked and thought, okay, we can do something that other people are willing to pay money for."
Kasesalu, Heinla, Tallinn. Three Estonian schoolboys, none of them yet twenty, who had learned to coax motion out of machines that resisted it. They would stay together, in one configuration or another, for the better part of two decades. Each of their small biographies points the same direction: talent formed under constraint, sold westward for currency that held its value.
The company they built, Bluemoon Software, was formally registered as a private Estonian firm on June 17, 1993. That year they released SkyRoads, a sequel to Kosmonaut, and a music editor called Sound Club, both as shareware. Then came a decade of whatever paid. They wrote firmware for an intelligent RAID hard-disk controller that made it into Byte magazine's "What's New International" section. They wrote a 6DOF 3D game called Stellar Xpress for a Texas virtual-reality gear company, bundled as a demo with its headsets. They built a secure financial transaction system, server and client both. They wrote a Windows scrolling driver called Tigma SuperScroll. They built software to run and maintain a GSM service center. They developed a 3D engine called Condor, the basis for games called The Art of Flying, Mars, and Thunder Brigade. In May 1997 they opened Bluemoon Interactive (UK) Ltd. in London and changed the product label to match.
"We jokingly called ourselves the 'computer games industry of Estonia,'" Tallinn said, "because we were really the only commercial computer games development studio in Estonia."
He had a phrase for this period, which he used in a talk: he described his life as "surfing the Moore's Law." The turning points, he noticed, coincided with what the exponential curve of cheaper computation made newly possible or newly pointless. When dedicated graphics cards arrived, games stopped being primarily a programming problem and became a storytelling and art-production problem. "Because we, being mostly good at programming, didn't have a good comparative advantage in this new world, we ended up exiting the computer games business."
Note the cool precision of that sentence. No nostalgia, no complaint about the industry selling out. A curve moved; their position on it changed; they left. This is how a physicist leaves a room.
He would apply the same reasoning, decades later, to a much steeper curve—and conclude that this time there might not be a next room to walk into.
Three Hundred and Thirty Dollars a Day
By 1999 Bluemoon was facing bankruptcy.
The founders did what talented, broke Eastern European programmers did at the end of the century: they sold their hours west. They took remote contract work for Tele2, the Swedish telecommunications company, at $330 a day each. The project was a web portal called Everyday.com. It was, in the flat language of the encyclopedia, "a commercial flop."
It was also where the Estonians met the two Scandinavians who would make them rich.
Niklas Zennström was the Swede who ran Tele2's operations in Denmark—a corporate operator with a talent for spotting where incumbents were overcharging. Janus Friis was the Dane he hired in 1996 to run customer support. Friis had no formal higher education; he had dropped out of high school and was working the help desk at CyberCity, one of Denmark's first internet service providers, when Zennström found him. Together they launched an ISP called get2net and then Everyday.com. Then they quit. In January 2000 Friis moved into Zennström's small apartment in Amsterdam, where the two of them set out to build a file-sharing company.
A dropout from the help desk, a telecom manager who wanted out, and three Estonians on a day rate. None of them had what anyone would call credentials. All of them had already failed at something.
The software they needed was written in Estonia. According to the account that has hardened into the record, Tallinn developed the FastTrack peer-to-peer technology and the Kazaa application for Zennström and Friis while "working as a stay-at-home father." The detail is easy to skip past and should not be. The protocol that would terrify the global recorded-music industry—that would, by 2003, make Kazaa the most downloaded piece of software on the internet—was being written by a young physicist at home in Tallinn with a small child. The Bluemoon team page from that era lists, in its last line, everything he apparently felt the world needed to know about his private life: "Is married and has a daughter."
Kazaa was ungovernable by design. There was no central server to subpoena, no Napster-style index that a judge could order switched off. Files moved from user to user to user. That architecture was the whole point, and it was also the whole problem. "Together with them, we first did the Kazaa file sharing application," Tallinn said, with Baltic understatement, "which got us into a bunch of legal trouble." Lawsuits from the American music and film industries followed; the business was eventually sold off.
What remained was the technical insight, and a team that had now been through bankruptcy, a flop, and a legal storm together. Sten Tamkivi, who later ran Skype's Estonian operation for eight years, put the arithmetic plainly: "From the outside, it's easy to see Skype as an overnight success… But in actuality, the same team had spent four to five years building various unsuccessful companies."
Four to five years. $330 a day. A portal no one used. This is the part of the story that usually gets cut.
A Phone With No Center
The idea, as the record has it, belongs to Janus Friis and Toivo Annus, the fourth Estonian on the team, who would go on to manage Skype's first office and its core network engineers: take the peer-to-peer machinery that had made Kazaa impossible to kill and use it to carry voice instead of MP3s. A telephone network with no switchboard. Calls hopping from one user's computer to another until they reached their destination.
The working name was "Sky peer-to-peer," shortened to "Skyper." The domain names were taken. They dropped the final "r."
Skype.com and Skype.net were registered in April 2003. An alpha version was tested that spring. The first public beta went live in Tallinn on August 29, 2003, built, as Estonian public television later noted, in "a fairly anonymous-looking, Soviet-era building in the Mustamäe district." About 10,000 people downloaded it on the first day. Within a few months it had a million users.
Tallinn and Kasesalu are listed as the software's original authors; Heinla and Annus as fellow builders. Skype itself was a Luxembourg company. The code was Estonian. A Dane designed the friendly interface. It was, the Estonian press would argue for twenty years afterward, the first truly global multinational startup nurtured in a city of that size.
Its genius was not that it was the first internet phone. It was not. "Skype had many competitors," recalled Martin Villig, who later co-founded Bolt. "Just as Google wasn't the first search engine." The advantage was that the thing simply worked, with no special modem, no adaptor, no configuration. Tamkivi described an internal standard they called the "granny test": could your grandmother download Skype unassisted and set it up? At the time, he noted, "most people had desktop computers with no headsets or built-in microphones." Suggesting everyone would soon talk through their computers was, as he put it, "a strange point in time" to suggest it. The bet was on where the curve was going.
By the summer of 2004, MIT Technology Review counted about 4.5 million registered users and reported that Skype had raised $18.8 million that March from Draper Fisher Jurvetson and Index Ventures. The reviewer placed an hour-long call from Riga to Smith College for nothing and reported that the voices arrived in "astonishingly clear audio." The magazine took care to note that the founders were not radicals out to destroy the telephone as an instrument of profit, but "simply ambitious businessmen intent on changing the economics of making a phone call."
Simply. There is nothing simple about breaking the economics of an industry, and Skype broke it for years. By 2009 it was considered the largest international voice carrier in the world by minutes. One estimate had it at 40 percent of the international calling market by 2014. In 2012, according to TeleGeography, its cross-border call growth alone was "more than twice that achieved by all international carriers in the world, combined." The telephone companies, those Goliaths, had been outmaneuvered by a piece of free software written in a Mustamäe office block by men who had been, five years earlier, broke.
It was also, in a way that matters to this story, a technology its own makers were not entirely sure they could constrain. Skype's architecture made it hard to wiretap; when the U.S. Federal Communications Commission extended its wiretapping statute to digital phone networks in 2006, Skype stated it did not plan to comply. Its partners in China ran a filtered version. When asked about that, Zennström gave an answer of perfectly polished corporate pragmatism: "Those are the regulations." Later, under Microsoft, it would appear in leaked documents about the NSA's PRISM program. When a Reddit user asked Tallinn about PRISM in 2013, he offered a careful, almost anthropological reply about the companies' communications staff: "what i've seen happen in PR departments is that they really want to avoid outright lying, but are OK with using careful wording and exotic definitions to make the meaning come out in certain light."
Hold that sentence too. Ten years later he would accuse an entire industry of something similar.
The Sixth Child
The money came fast and then the company slipped away from its makers.
On September 12, 2005, eBay agreed to buy Skype for approximately $2.5 billion in cash and stock up front plus performance-based consideration, roughly $2.6 billion in total. Tallinn sold his shares to eBay. This turned out to be either prudent or slightly unlucky, depending on how you count. Two years later eBay took a $1.4 billion impairment on Skype, effectively admitting it had overpaid. In 2009 it sold 65 percent of the business to an investor group led by Silver Lake at a valuation of about $2.75 billion. On May 10, 2011, Microsoft agreed to buy the whole thing for $8.5 billion—32 times Skype's operating profits, a price the Financial Times said raised fears of a new tech bubble.
Asked in 2013 how much of the $8.5 billion he'd taken home tax-free, Tallinn answered with the deadpan of a man correcting a decimal point: "nothing, because i sold my shares to EBAY, not microsoft."
The scene that best captures his relationship to Skype in its afterlife comes from a 2013 Ars Technica piece by the Estonian journalist Toivo Tänavsuu, written for Skype's tenth birthday. Tallinn arrives at an open-air restaurant in the old town of Tallinn—a man named for his city, sitting in it—takes off a pair of blue sunglasses, finds a seat, and says: "I don't care about Skype!"
"The technology is 10 years old—that's an eternity when it comes to the Internet Age," he continued. "Besides, I have more important things going on now."
At the time he had five children. He called Skype his sixth.
Is that a contradiction? A father who doesn't care about one of his children? Not quite. It is the attitude of a programmer toward code that has shipped: it belongs to its users now, and to its owners, and the author's job is done. By then the company employed hundreds in Tallinn and Tartu—44 percent of the entire division's staff was in Estonia as of 2011—and its alumni were scattering to found the next generation of Estonian companies. Taavet Hinrikus, Skype's very first employee, went on to build TransferWise. "By creating, having and maintaining Skype in Tallinn," Hinrikus said, "we gained a great insight on how to launch a great global product and it created a feeling that we can create big things in a small place." Tamkivi went further: "We came to see being small as an advantage."
The four Estonian engineers pooled their proceeds into an investment company called Ambient Sound Investments. Tallinn would eventually be recognized by the Estonian state with the Order of the White Star. One of his old memberships lists him as holding the world record for the most software downloads, at nearly 500 million, between Kazaa and Skype.
In 2010, at a Singularity Summit event, he gave a talk whose opening a listener named Nathan Labenz never forgot. Tallinn said he found himself in the unusual position of trying to figure out what to do next, knowing he had just wrapped up what was probably going to be the most successful project he would ever do.
He was, as it turned out, wrong about that. But the more important more things going on were already underway. They had begun at the franchise eatery by the freeway.
Messages From Behind the Curtain
When Tallinn finally wrote about that 2009 meeting at length, he framed it inside a larger idea. He called the essay "Dissident Messages," and he began not with AI but with his childhood.
Following Norbert Wiener, the mathematician who saw the world in terms of signals and noise, Tallinn treated history as the transmission of messages. The first message he described was the Soviet occupation itself: the official story of a voluntary union, broadcast relentlessly, and the counter-message—the dissident one—passing quietly through Estonian society by other channels, through foreign radio like the Voice of America, until it surfaced as the Singing Revolution and the country walked out of the empire. A message that seemed fringe and dangerous to repeat, then suddenly obvious.
He was telling his readers how to hear Yudkowsky.
The message itself, as Tallinn summarized what struck him, was simple and enormous: "The overall idea that caught my attention that I never had thought about was that we are seeing the end of an era during which the human brain has been the main shaper of the future."
He took the next step that rich people who become convinced of something usually do not take. He went to work on it full-time. "By the following year, I had dedicated my time to existential risk mitigation with a focus on AI." In a 2013 Reddit thread he gave a one-sentence autobiography that reads like an optimizer's report: he had "figured out that to further maximise my causal impact i should join the few good people who are trying to reduce existential risks." He added, almost apologetically, that "all important topics seem to bottom out in philosophy!"
— Jaan Tallinn, Edge.org, 2015I was born behind the Iron Curtain, in Soviet-occupied Estonia, and looked forward to a pretty bleak life in some scientific institute trying to figure out how to kill more Americans.
The first problem was respectability. Talking about machine superintelligence in 2010 got you filed with the people who talk about alien abductions. So Tallinn did what a physicist would do: he went and found other physicists. "If you look around in the so-called existential risk ecosystem that I support," he observed, "there's, I would say, an over-representation of physicists, because physics helps you to see the world in a neutral manner." In 2012, with the philosopher Huw Price and the astronomer Martin Rees—Britain's Astronomer Royal, a man whose credibility could not be questioned—he co-founded the Centre for the Study of Existential Risk at the University of Cambridge, putting in about $200,000 to start it. Two years later, in March 2014, he co-founded the Future of Life Institute in the other Cambridge, Massachusetts, with the MIT cosmologist Max Tegmark, the physicist Anthony Aguirre, and others. (Tegmark's later book, Life 3.0, would become one of the movement's main public texts.)
The respectability strategy worked faster than anyone expected. In January 2015 FLI published an open letter calling for research to ensure AI systems were "robust and beneficial," with the line "our AI systems must do what we want them to do," and gathered signatures from Stephen Hawking and Elon Musk. Musk gave FLI $10 million that year. Morgan Freeman, of all people, sat on the institute's scientific advisory board. "Our first goal," Tallinn said of this period, "is to bring more respectability to these issues."
He funded the Machine Intelligence Research Institute, Yudkowsky's organization, which by 2026 credited him with more than $1.6 million in total contributions. He gave the University of Tartu €54,000 for a scholarship that, in its first round in January 2014, sent five science students to a workshop run by the Center for Applied Rationality in Melbourne. "The present century is obviously largely designed by the people of science, i.e. by researchers and programmers," he explained. He joined the Board of Sponsors of the Bulletin of the Atomic Scientists, the publication that keeps the Doomsday Clock—a fitting home for a man who had grown up expecting to work on weapons.
His method, when he described it, was not the method of a zealot. It was the method of a debugger. "My general approach is to basically look at when people are making arguments and see whether I can find a bug in them," he told an Effective Altruism audience in 2018. "If I can't find a bug, I just say, 'Okay, here's some probability mass for you.'" The cost of this generosity was certainty; as more arguments arrived, his views diffused. "I'm missing the days when it was just one fairly crisp idea of what might go wrong."
He gave a talk in these years—at Tohoku University in 2017 and elsewhere—built around a single image, and it is the clearest window into how he thought. Imagine, he said, a giant spaceship large enough to carry every human being on Earth. Construction is far from finished but boarding has begun, the children already in their seats. The project consumes billions of dollars and millions of hours a year. And a small group of people is trying to point out, over the noise of construction, that the ship has no steering mechanism. The engineers wave them off. It will never take off; steering is trivial; steering is impossible; worrying about payload safety is selfish and backward. All the factions, he noticed, had two things in common: they were making excuses why nothing needed to change, and they were covering up a grossly embarrassing blunder.
The metaphor is funny if you let it be. It is also unsettling, because of who is telling it. He had spent the previous decade building the engines.
Money That Cares
In 2011 Tallinn invested in a small London company called DeepMind.
It had been founded the previous November by Demis Hassabis, Shane Legg, and Mustafa Suleyman. Hassabis and Legg had met at the Gatsby Computational Neuroscience Unit at University College London; their goal, at a time when the phrase sounded like science fiction, was a general-purpose artificial intelligence. They taught it, first, to play Atari games—Breakout, Pong, Space Invaders—with no programmed knowledge of the rules, only raw pixels and a score. After a period of flailing, the system would learn what worked. Then it would become better at the games than any human. Hassabis, a former games developer, had built a research lab on the logic of play. So had Tallinn, in a sense, twenty years earlier, at the slow end of the curve.
Tallinn put in money and joined the board as an investor director, alongside Founders Fund and Horizons Ventures, Peter Thiel and Elon Musk. His reason was the reverse of every other investor's. He was not there for the upside. He was there to have a seat in the room. "I have continued to invest in various AI companies in order to have a voice of concern from the inside," he wrote in Newsweek in 2023. CNBC put it more bluntly in 2020: the early investments were partly "so that he can keep tabs on what they're doing."
He gave the philosophy a name, which he repeated for years in different words: "My philosophy has been that I want to displace money that doesn't care." On the Future of Life Institute's podcast he was more specific about the enemy: "the profit maximizing money."
— Jaan Tallinn, to Semafor, April 2023My philosophy has been that I want to displace money that doesn't care.
The theory is elegant. Every cap table has a fixed number of seats. If a safety-minded investor occupies one, a growth-at-all-costs investor does not. Influence follows ownership; conscience can be bought in at par. He told an Estonian business paper in 2021 that he had bought stakes in some companies precisely to stand in the way of developments he thought could become dangerous, and added, as if it were an unremarkable point of financial ethics, that as a man investing his own money he was under no obligation to maximize it.
The first real test came in January 2014. DeepMind needed a well-funded parent to pay for computing and salaries; Facebook and Google were both courting it. On January 26, 2014, Google confirmed the acquisition, at a price reported to be somewhere between $400 million and $650 million. One condition of the deal was an AI ethics board. Tallinn was part of the conversations that produced that promise. "While that ultimately failed," he wrote nine years later, "at the time it felt like progress." The board's membership was never disclosed. It became, in the press, a mystery and then a footnote.
In the decade that followed, the lab he had backed out of fear did things that made his fear look prescient and his money look brilliant at the same time. AlphaGo beat Lee Sedol four games to one in March 2016. AlphaGo Zero, trained with no human games at all, beat its predecessor a hundred games to none. AlphaFold largely solved protein structure prediction, and in October 2024 Hassabis shared half a Nobel Prize in Chemistry for it. The zero in those names haunted Tallinn in a particular technical way. "Zero and alpha zero meant zero data," he told Hsu. "So you basically run fully on synthetic data." The machine no longer needed us even as a source of examples.
There were other bets, dozens of them. By 2015 he was investing in 34 startups at once—Lingvist, which built a machine tutor for language learning; Nightingale, which automated parts of behavioral therapy for children with autism; Fleep, a messaging app in which he wanted readers to rate the urgency of messages so that habitual alarmists would lose credibility, a reputation system for cried wolves. He co-founded MetaMed, a personalized medical research firm backed by Peter Thiel; its failure, he said later, taught him the importance of feedback loops. He became, through a vehicle called Metaplanet founded in 2011, a prolific angel, and later handed most of the legwork to a dedicated team. "Time is my most precious resource," he said, "so I try to co-invest with other people."
Some of the bets were not bets at all. In January 2018 he gave Faculty, a London AI company in which Metaplanet held just under 9 percent, 350 units of Ether, then worth about $434,000; in March 2020 he gave it 50 Bitcoin, worth about $316,000—a gift, not an investment, to fund work on the unglamorous near-term problems of algorithmic bias and transparency. A man who thinks in centuries paying, in cryptocurrency, for someone to worry about this quarter.
And some of the counterparties were not what they seemed. According to sources cited by the Wall Street Journal and Semafor, Tallinn lent about $100 million to Alameda Research, the trading firm co-founded by Sam Bankman-Fried, then recalled most of the loan in 2018—years before the collapse of FTX made the name a byword. Semafor's headline about that early money: "Charity-linked money launched Sam Bankman-Fried's empire." Money that cares can still be lent to the wrong people. He pulled it back early. It is the kind of outcome that gets remembered, if at all, as good judgment rather than luck, and he would probably assign it some probability mass of each.
Plan A
Through MIRI, Tallinn met many of the people who would later build the frontier labs. One of them was Dario Amodei, a physicist turned AI researcher who had gone to work at OpenAI—itself an organization Tallinn had backed in its early days, conceived as a counterweight to Google and DeepMind. When Amodei and a group of colleagues left OpenAI in 2021, primarily, it was reported, out of concern that OpenAI was not focused enough on safety, they founded a company called Anthropic.
Tallinn led its Series A. On May 28, 2021, Anthropic announced it had raised $124 million. One later account put the post-money valuation at $623 million, a figure it described as contrarian for a lab that had just been founded. Four years later, in September 2025, Anthropic raised $13 billion at a $183 billion valuation. An Estonian newspaper headlined its year-end profile of him "Jaan Tallinn: becoming a billionaire with the help of AI."
Here the story bends. Tallinn had told Hsu that his rule for AI investments was to have a voice and displace uncaring money, but "the main constraint that I tried to satisfy is to not accelerate those companies. And so I never kind of like a major investor. I'm always like a small investor." Leading a $124 million Series A is not what most people mean by small. He knew it. In April 2023 he told Semafor's Reed Albergotti that he was not sure he had made the right decision. "On the one hand, it's great to have this safety-focused thing," he said. "On the other hand, this is proliferation."
He praised Anthropic as the most safety-conscious of the labs. And then he said the most honest thing a large shareholder has probably ever said about his own holding: "that doesn't change the fact that they're dealing with dangerous stuff and I'm not sure if they should be. I'm not sure if anyone should be."
— Jaan Tallinn on Anthropic, to Semafor, April 2023That doesn't change the fact that they're dealing with dangerous stuff and I'm not sure if they should be. I'm not sure if anyone should be.
The trigger for that interview was a letter. In March 2023, the Future of Life Institute—his institute—published "Pause Giant AI Experiments," calling for "all AI labs to immediately pause for at least 6 months the training of AI systems more powerful than GPT-4." Elon Musk signed. Steve Wozniak signed. Tallinn signed. Anthropic did not. Its co-founder Jack Clark explained that the company did not sign petitions, as a matter of policy, and the company issued a statement welcoming debate about "different approaches." A month later Anthropic accepted a $300 million investment from Google.
"Plan A failed," Tallinn said. "There is a dissonance between privately being concerned and then publicly trying to avoid any steps that would address the issue."
It is worth putting that sentence beside the one from 2013, about PR departments that "really want to avoid outright lying, but are OK with using careful wording." The concern was real, he believed. The people were decent; he had said so himself on Edge.org in 2015, with conspicuous calm: "I know that they are well-meaning and people with good integrity." The problem was structural. Inside a company, he wrote in Newsweek, there were always sympathetic people, "but there's only so much they can do once they are within the constraints of that company." The seat at the table bought a voice. It did not buy a veto.
So the founder of a peer-to-peer network—the architecture designed specifically so that no one could switch it off—turned, in his fifties, into an advocate for switches. In May 2023 he signed the Center for AI Safety's one-line statement that "mitigating the risk of extinction from AI should be a global priority alongside other societal-scale risks such as pandemics and nuclear war." That year he said he had yet to meet anyone working at the labs who put the risk of the next model "blowing up the planet" below 1 percent. In October 2025 he signed the Future of Life Institute's call for "a prohibition on the development of superintelligence, not lifted before there is broad scientific consensus that it will be done safely and controllably, and strong public buy-in"—a list on which his name sat alongside, among others, Prince Harry and Meghan Markle.
There is something almost comic in that last pairing, and something almost tragic too. The engineer from Mustamäe and the Duke of Sussex, on the same page, asking the same companies to stop. He was not on their page by accident. He had spent fifteen years building the page.
The Hum in the Cellar
Listen to Tallinn describe a frontier training run and you start to hear a horror story told in the vocabulary of systems administration.
There are, he told Hsu in May 2024, two frontiers. One is deployment—where people figure out where to plug AI into the world. That one he finds almost reassuring, because "there are a bunch of experiences that we have regulating and constraining new technology." The other is pre-training: "those months-long periods during which megawatts of electricity and" (he corrected himself mid-sentence) "terabytes of data are being poured into this like black box. And there's that just like hums there in the cellar or whatever the data center is."
What happens in the cellar? Nobody can say for certain, and that is his point. The training data contains everything humans have written about computer security, including how to break it. "I used GPT 4 to help me to configure my own firewall," he said. "It just knows this thing. It just doesn't connect the dots yet." He was careful to distinguish the model itself from the characters it can simulate: an agent conjured inside the model, prompted by some sequence in its training text, might start treating "the compute substrate that is being instantiated in as its environment." He called this "pure speculation." His point was narrower and harder to wave away: "we do not have guarantees that something like that is not happening."
His language for what the labs are making has the flatness of a man who has stopped looking for a gentler word. "Now we human comparable minds that you can run on a USB stick. Which is a very, very weird situation." And: "AI is building aliens. Building digital aliens." He compared the current paradigm—grow a mind by pouring compute and data into it and see what emerges—to gain-of-function research in biology. To an interviewer at a European venture firm he reduced his whole concern to an image from primatology: "creating AIs that we will not be able to control, much like chimpanzees cannot control humans."
He keeps a short list of policy priorities on his personal website, and it reads like a programmer's spec sheet for civilization. Six items. Certification of the data centers where frontier models are trained. Speed limits on AI inference, because a machine thinking a million times faster than people would make humans "just plants" by comparison. Liability laws to send blame back to wherever a problem is easiest to fix. Labeling, so machine output can't be mistaken for human. Veto committees, broadly representative bodies that would need to approve risky projects. And global off switches.
The off switch is where his engineering instincts and his biology reading meet. Synthetic biologists, he pointed out, sometimes make engineered organisms depend on a chemical that does not occur in nature, so that they die if they escape. Do the same with chips, he suggested: give GPUs and TPUs a secure enclave that expects a stream of cryptographically signed tokens, and if the tokens stop, the chip shuts the rest of itself down. "I think that's totally implementable," he said. "And I think it's a responsible thing to do." A machine that cannot keep running without permission from outside.
He has no illusions that safety research is a clean good. "One lesson that I learned over the last decade or so," he told Hsu, "is that so-called AI safety and AI alignment is like a double-edged sword, because all advances in safety pretty much are immediately put into AI capabilities." Understand a model better and you can also make it cheaper and stronger. The steering mechanism doubles as a second engine.
And yet he keeps his eye on the human pieces of the board, with occasional dry humor. Hsu, a self-described Cold War kid, asked whether it would help to commission the director of Dune to dramatize the Butlerian Jihad, the novel's backstory in which humanity outlaws thinking machines. Tallinn didn't think ordinary people needed convincing; the polls, he said, showed them asking "wait a minute, why are we doing this?" The resistance, in his experience, came from "a subclass of intellectuals, who just want to say smart things," people who follow the heuristic of not saying crazy-sounding things. And on the Terminator films, which the AI safety community had spent a decade resenting as the stock photograph on every article about their work: "the movie is not actually bad." A realistic AI disaster, he observed, would not be interesting to humans. Hollywood had done the best it could under the constraint of needing an audience.
Pressed for a timetable, he does not offer one so much as admit that the one he had is gone. "There was this period during which these concerns seemed fairly remote," he said in a 2024 documentary interview. "Most people in the field, in the AI safety field thought that we probably still had until the middle of the century or so before we really need to start worrying. It is now becoming increasingly clear that we will not have until the middle of the century. Possibly we will only have a few years left now." In 2024 he put it more starkly still: "there is a substantial chance that humanity might destroy itself with AI over the next decade or two, if we do not begin approaching AI development with a great deal more care and caution."
He has a recurring word for the time remaining: runway. Every new release of a more capable model, he told Hsu, will "eat into our runway as a civilization." But each also hardens public resolve, the way ChatGPT did; he noted that the legislation, executive orders, and international bodies had all appeared downstream of that one product. The runway shortens; the will to brake grows. Which wins is the only question.
When pushed on what any of it is for, he reaches for a physicist's frame and a hobbyist's vocabulary. "In the end," he said, "the game is what kind of configurations of atoms are reachable from here and do they contain human shaped objects." Five years earlier, asked what the good outcome looked like, he had said only: "I like to think in terms of fun; I want to be more fun than less fun." When he spoke to Harvard undergraduates in 2013, after a talk that wove together Skype, existential risk, and what he called, tongue in cheek, a "sermon," he ended by arguing that the one thing indispensable to building robust companies and keeping the species safe was having fun.
That is not a contradiction either, quite. It is a man who has spent his life working on games and protocols, and does not want the future to be boring, and does not want it to be empty.
In September 2026, Estonia's leading business paper put him at the top of its rich list, with an estimated net worth of €10.3 billion. A few months earlier, Time had named him to its list of the world's most influential philanthropists. The fortune and the philanthropy grew from the same fear, compounding at the same rate. In March 2026, at an Estonian employers' conference, sharing a stage with Taavet Hinrikus, Skype's first employee, he warned that AI would reshape the labor market not over decades but over the next few years, and not in one great wave of layoffs but quietly. He still lives in Tallinn. He still conducts a remarkable number of interviews over Skype.
Or did.
Remaining Balance
On February 28, 2025, Microsoft announced that Skype would be retired. On May 5, 2025, it was.
The end was undramatic. Daily users had fallen to 36 million by 2023, eaten first by Zoom during the pandemic and then by Microsoft's own Teams. The peer-to-peer architecture that had made Skype ungovernable had been dismantled years before: by May 2012 the network ran on Microsoft-hosted supernodes, and by 2017 it was a conventional centralized service on Azure, with a switch in Redmond that could, in fact, be flipped. The website began redirecting to Teams. Users were told to export their chat histories by January 2026; in December 2025 Microsoft extended the deadline to June 2026.
The sixth child was put down gracefully, on schedule, by the people who owned it, with a deadline extension and an export tool. It is the only technology in this story that anybody has ever successfully turned off.
One small part of it is still running. Microsoft kept a web page called the Skype Dial Pad alive for people who had paid for calling credit before the shutdown. They can still use it to ring landlines and mobile phones around the world, the way a student in Riga once rang her sister in Massachusetts. The balance can be spent down, call by call, minute by minute.
No new credit can be purchased.
Part IIThe Playbook
What follows is an attempt to extract operating principles from a career that runs, unusually, in two directions at once: a builder who learned to ship software no one could switch off, and an investor who now spends his fortune trying to make sure the next thing can be switched off. Some of these principles are tactical, drawn from Bluemoon's decade of contract work and Skype's ascent. Others are stranger, drawn from Tallinn's attempt to buy a conscience into the capital structure of the AI industry and his frank admission that it didn't work. All of them are grounded in what he actually did, and in what he has said about doing it.
The Tallinn Playbook
- 1.Locate yourself on the curve, and leave when it moves.
- 2.Count the failed companies as tuition, not detours.
- 3.Repurpose an architecture across domains.
- 4.Pass the granny test.
- 5.Take the exit that funds the next mission.
- 6.Debug arguments; assign probability mass.
- 7.Borrow credibility before you spend conviction.
- 8.Displace money that doesn't care.
- 9.Say out loud when Plan A has failed.
- 10.Ask for invariants, not intentions.
- 11.Treat your time as the scarce asset.
- 12.Keep fun in the objective function.
J
Two Arcs, One Career
Tallinn's life divides into building ungovernable networks and then arguing for governable machines.
1986
At fourteen, starts programming 8-bit Intel 8080 school computers with classmate Priit Kasesalu.
1989
Kosmonaut, built with Kasesalu and Ahti Heinla, becomes the first Estonian game sold abroad; earns $5,000.
1993
Bluemoon Software registered on June 17; SkyRoads and Sound Club released as shareware.
1996
BSc in theoretical physics, University of Tartu; thesis on interstellar travel via spacetime warps.
1999
Bluemoon nears bankruptcy; founders take Tele2 contract work at $330/day each.
2003
Kazaa is the world's most downloaded software; Skype public beta launches August 29.
2005
eBay buys Skype for ~$2.6B; Tallinn sells his shares.
2009
Four-hour meeting with Eliezer Yudkowsky in March.
2011
Invests in DeepMind; founds Metaplanet.
2012
Co-founds CSER at Cambridge with Huw Price and Martin Rees.
2014
Google acquires DeepMind; Tallinn co-founds the Future of Life Institute.
2021
Leads Anthropic's $124M Series A.
2023
Signs pause letter and extinction-risk statement; tells Semafor "Plan A failed."
2025
Skype retired May 5; signs call for a prohibition on superintelligence.
2026
Tops Äripäev's Estonian rich list at €10.3B.
Principle 1
Locate yourself on the curve, and leave when it moves
Tallinn's own summary of his first career is that he spent it "surfing the Moore's Law." What makes the phrase useful is not the surfing but the exits. When dedicated graphics cards arrived, the value in games migrated from programming to art and story. Bluemoon was a team of programmers. "Being mostly good at programming," he said, they "didn't have a good comparative advantage in this new world," so they left games entirely and went into internet software, where programming still was the scarce input.
Most founders treat their industry as their identity. Tallinn treated his as a position on a curve that would keep moving whether he liked it or not. The same frame informs his AI thinking: his central forecasting question, he told CNBC, is "how strongly and how exactly will AI development feed back to AI development," the slope of the curve rather than its present height. A team's edge is a function of where the exponential is, not where it was when the team formed.
The discipline is to name your comparative advantage precisely enough that you can tell when it has stopped being scarce. Bluemoon could say in one word what it was good at. That clarity is what let it walk out of a dying business before the business walked out on it.
Tactic: Write down, in one noun, the input your team supplies that the market can't easily get elsewhere, and re-check every year whether technology has made that input cheap.
Principle 2
Count the failed companies as tuition, not detours
The overnight success of Skype had a long night before it. Bluemoon nearly went bankrupt in 1999. Everyday.com was a flop. Kazaa was buried in lawsuits. Sten Tamkivi, who ran Skype Estonia, insisted on the arithmetic: "the same team had spent four to five years building various unsuccessful companies." The RAID firmware, the GSM center software, the financial transaction system, the scrolling driver—none of them were glamorous, but each paid the bills and each taught the team something about shipping software for strangers.
The pattern underneath is that the failures were team-preserving. Kasesalu, Heinla, and Tallinn stayed together from 1986 to Skype. When they took $330 a day from Tele2, they took it together, and that contract is where they met Zennström and Friis. A failure that keeps the right people in the same room is an investment in the next attempt, not a write-off.
Tallinn himself absorbed the lesson more slowly in later ventures. MetaMed's demise, he said, taught him "the importance of feedback loops"—the same lesson, at a later date, in a different industry.
Tactic: When a venture fails, measure what survived—the team, the codebase, the relationships—and choose your next contract work for its proximity to people you'd want to build with.
Principle 3
Repurpose an architecture across domains
Skype was not a new technology; it was an old architecture pointed at a new payload. The peer-to-peer machinery behind Kazaa, which moved MP3s from user to user without a central server, was redirected to carry voice. The name itself records the move: "Sky peer-to-peer." The team didn't have to invent decentralized routing; they had to notice that the thing that made file sharing cheap could make telephony cheap.
The economic consequence was enormous. As MIT Technology Review noted in 2004, competitors routing calls over the internet still depended on centralized servers and had to charge accordingly; Skype "simply used the internet" and had no complicated infrastructure to build. The architecture was the business model. By 2009 Skype carried more international minutes than any carrier in the world.
The general move is to treat your hardest-won technical capability as a solution looking for additional problems. The domain where you built it is merely the first customer.
Tactic: List the three hardest technical problems your team has solved, then ask which other industries currently pay a premium because they haven't solved them.
Principle 4
Pass the granny test
Skype was not the first internet phone, and its early competitors were not technically incompetent. Its advantage, as Bolt co-founder Martin Villig put it, was that "it was made so technologically simple that anyone could install it and use it." The internal benchmark was the "granny test": could your grandmother download Skype unassisted and set it up?
This mattered more because the bet was ahead of the hardware. Tamkivi recalled that most people at the time had desktops with no headsets or built-in microphones. When you are asking users to change behavior before the world is ready, every unit of friction you remove is a unit of adoption you gain. Ten thousand downloads on day one and a million within months came from a product that had no onboarding problem.
The insight generalizes: technical depth wins markets only when it is invisible. Kazaa's and Skype's protocols were sophisticated; the user experience was a button.
Tactic: Before launch, hand the product to the least technical person you know, say nothing, and fix everything that makes them ask a question.
Principle 5
Take the exit that funds the next mission
Tallinn sold his Skype shares to eBay in 2005 at a roughly $2.6 billion valuation. Six years later Microsoft paid $8.5 billion. Asked in 2013 how much of the latter he received, he answered: "nothing, because i sold my shares to EBAY, not microsoft." He did not sound bitter.
He had reasons not to be. The eBay exit gave him liquidity at the moment he needed freedom to look for the next problem—the moment that led to the 2009 meeting with Yudkowsky and, within two years, the DeepMind stake. Holding on would have meant riding eBay's $1.4 billion impairment and the Joltid licensing dispute before Microsoft's rescue. More important, he had stopped thinking of Skype as his project; by 2013 he was saying "I don't care about Skype!" and that he had "more important things going on now."
The principle is not "sell early." It is that an exit should be evaluated against what the capital and attention will do next, not only against the maximum price the asset might eventually fetch.
Tactic: Before negotiating an exit, write down what you'd do with the proceeds and freed time, and value the exit against that plan rather than against a hypothetical peak.
Principle 6
Debug arguments; assign probability mass
Tallinn's epistemics come straight from programming. "One valuable thing as a programmer is that you understand when you understand and you understand when you don't understand," he said, "because if I understand something in a way that I could describe to a computer, and then it works after that, I know that I really understood it." Applied to arguments, this becomes a procedure: look for the bug. "If I can't find a bug, I just say, 'Okay, here's some probability mass for you.'"
It is how a successful founder ended up taking a blogger's warning seriously in 2009 when most people of his stature would have pattern-matched it to science fiction. It is also why his views became less certain over time rather than more: as more credible arguments arrived, each got a share of probability. He admitted he missed "the days when it was just one fairly crisp idea." He recently funded a project in Berkeley explicitly to reconcile competing hypotheses.
He pairs this with first-principles reasoning, which he borrowed from Elon Musk's distinction between metaphorical and first-principles thinking. For risks that have never happened, he argued, analogy fails: "you have to engage in this slow and laborious process of listening to the arguments and not pattern-matching them to things that you think might be relevant."
Tactic: When an argument sounds crazy, try to find the specific step that fails; if you can't, give it a nonzero probability in writing and revisit it.
Principle 7
Borrow credibility before you spend conviction
Tallinn's first stated goal for the existential-risk institutions was not research. It was respectability. "Our first goal is to bring more respectability to these issues." So he co-founded CSER with Huw Price and the Astronomer Royal, Martin Rees, and FLI with a group of MIT and university physicists. He pointed to the over-representation of physicists as a feature: they helped "see the world in a neutral manner."
The payoff was rapid. Within a year of FLI's founding, its open letter carried Stephen Hawking's and Elon Musk's names, Musk had donated $10 million, and the topic had moved, in Tallinn's phrase, inside the Overton window: "You can actually publicly talk about it without being ostracized."
A wealthy outsider with an unpopular idea who funds his own advocacy is easy to dismiss. A wealthy outsider who funds institutions led by people whose credentials cannot be questioned is much harder to ignore. The money buys the platform; the borrowed credibility makes the message land.
Tactic: If you hold an unpopular conviction, fund institutions led by people whose reputations are stronger than yours, and let them be the face of it.
Principle 8
Displace money that doesn't care
Tallinn's investment thesis inverts the venture model. Most investors buy equity to capture returns; he bought it to occupy seats. "My philosophy has been that I want to displace money that doesn't care," he told Semafor. On the FLI podcast he named the target: "the profit maximizing money." He told Ärileht he had bought stakes in some firms simply to impede dangerous developments, and that as an investor of his own money he was not obliged to maximize it.
The mechanism has real effects. It gave him a board seat at DeepMind and a role in the negotiations that produced the promise of an ethics board at the Google sale. It made him the lead investor at Anthropic rather than leaving that slot to someone indifferent to safety. And because his returns were irrelevant to his purpose, he could make gifts where equity would distort, as with the crypto donations to Faculty for near-term AI ethics work.
But the thesis has a built-in hazard: displacing uncaring money also means supplying capital, and capital accelerates. That is why he imposed a second constraint on himself—to be "always like a small investor"—and why leading Anthropic's round troubled him.
$
Conventional Venture vs. Displacement Investing
How Tallinn's stated approach diverges from standard practice.
| Dimension | Conventional investor | Tallinn's approach | Outcome |
|---|---|---|---|
| Purpose of equity | Capture returns | Occupy a seat a profit-maximizer would have held | Achieved |
| Position size | Concentrate in winners | Stay small to avoid accelerating | Broken at Anthropic |
| Governance ask | Information rights | Ethics board as condition of DeepMind sale | Promised, then failed |
| Vehicle for non-financial goals | Separate foundation | Gifts to portfolio companies (Faculty, in crypto) | Experimental |
| Exit expectation | Maximize multiple | No obligation to maximize | Billions anyway |
Tactic: If you invest for influence, decide in advance the maximum share of a round you'll take so your capital doesn't become the accelerant you meant to restrain.
Principle 9
Say out loud when Plan A has failed
The rarest thing in this profile is not the fortune or the foresight. It is a billionaire telling a reporter that his strategy did not work. "Plan A failed," Tallinn told Semafor in April 2023, after the labs he had backed declined to sign his institute's pause letter. "There is a dissonance between privately being concerned and then publicly trying to avoid any steps that would address the issue." Of the Anthropic investment, he said he was not sure he had made the right decision.
The candor is functional. Admitting failure let him change approach rather than defend a sunk cost. He told Hsu in 2024 that he was "in the midst of switching my strategy" from funding research at a distance to "more direct action": backing competent groups working on provable safety and legislation, sitting on the UN AI advisory body, and publishing concrete policy priorities.
Founders and investors often treat public admission of failure as reputational suicide. In Tallinn's case it increased his credibility, because it showed his warnings were not self-interested positioning.
Tactic: Schedule a written review of each major strategy at a fixed date, and if it has failed, say so publicly and state what you will do instead.
Principle 10
Ask for invariants, not intentions
After a decade of trusting well-meaning people inside companies, Tallinn concluded that good intentions don't survive organizational constraints. His proposals now ask for things that hold regardless of anyone's goodwill—what programmers call invariants. "That would also give us some kind of invariant that we can reason about," he said of chip-level off switches, "but currently just very little invariance."
His six priorities read as a spec: certified data centers, speed limits on inference, liability that propagates to where problems are cheapest to fix, mandatory labeling, veto committees, and global off switches. His most concrete idea borrows from synthetic biology's practice of making organisms depend on a chemical absent in nature: GPUs with secure enclaves that require a stream of signed tokens and shut down when it stops.
The engineering lesson applies far beyond AI. Commitments that depend on someone remembering to keep them are weaker than mechanisms that enforce themselves. The man who built software no one could switch off knows exactly how hard it is to retrofit a switch.
Tactic: For any risk you care about, replace "we will be careful" with a mechanism that fails safe automatically if no one intervenes.
Principle 11
Treat your time as the scarce asset
"Time is my most precious resource," Tallinn said in 2015, when he was investing in 34 startups at once, "so I try to co-invest with other people. If I find people I can trust to handle the due diligence, I can save time." By 2020 he described his professional life as roughly 50/50 between investing and philanthropy, traveling about 40 percent of the time, and said he had hired someone to do most of the investing legwork "to grant me more time for actually saving the world." By 2024 he had delegated investing to a dedicated full-time team.
The allocation follows his priorities. Investing produces money; money is fungible; others can manage it. The thing no one else can supply is his attention on the problem he thinks matters most. He spends the rest of his time, he said, maximizing conversations "with people whose ability to think I respect."
For founders who become investors, the trap is that dealflow expands to fill every hour. Tallinn's answer was to treat deal evaluation as delegable and judgment about priorities as not.
Tactic: Identify the one activity only you can do, then delegate or co-invest away everything that merely requires competence.
Principle 12
Keep fun in the objective function
For a man who talks about extinction, Tallinn is notably unwilling to let gravity flatten his goals. Asked in 2019 what a good future would look like, he said: "I like to think in terms of fun; I want to be more fun than less fun." He warned against locking in present values, because "we know our values have progressed over time." At Harvard in 2013, he argued that indispensable to building robust businesses and keeping the species safe was having fun.
This is not frivolity. It reflects his origins: a games developer who got hooked on programming because it was play under constraint, and who now worries about a technology developed, at DeepMind, by teaching machines to play Atari. It also keeps the mission pointed at something worth saving. "I really care about humans," he said, "and I think it's okay to care about humans."
Missions built entirely on fear tend toward burnout or zealotry. Missions that keep a positive objective—a future that is more fun rather than merely less dead—tend to keep their people.
Tactic: State the positive outcome your work protects in a single plain sentence, and revisit it whenever the defensive work starts to feel like the whole job.
Free playbook
Get The Business Model Playbook
58 business models, one visual page each: how the money flows, the metrics that matter, and who runs it. Free when you join the Faster Than Normal email.
Free. No spam. Unsubscribe anytime.
Part IIIQuotes and Maxims
In their words
— Jaan Tallinn, on Eliezer Yudkowsky's writing, Reddit AMA, 2013The overall idea that caught my attention that I never had thought about was that we are seeing the end of an era during which the human brain has been the main shaper of the future.
— Jaan Tallinn, to Semafor, April 2023Plan A failed. There is a dissonance between privately being concerned and then publicly trying to avoid any steps that would address the issue.
— Jaan Tallinn, keynote at the Vienna Conference on Autonomous Weapons Systems, 2024AI can be a tool, but tools are only valuable as long as they are predictable and can be controlled.
— Jaan Tallinn, Edge.org, 2015Because we, being mostly good at programming, didn't have a good comparative advantage in this new world, we ended up exiting the computer games business and going into Internet programming.
— Jaan Tallinn, Manifold podcast with Steve Hsu, 2024In the end the game is what kind of configurations of atoms are reachable from here and do they contain human shaped objects.
Maxims
- Know your slope. Your advantage is a position on a moving curve; leave before the curve leaves you.
- Failure that keeps the team is tuition. Bankruptcy, a flop, and a lawsuit were the prerequisites for Skype, not interruptions to it.
- Point old architecture at new payloads. The protocol that moved songs moved voices; your best tool has more than one customer.
- Invisible depth wins. Sophistication belongs in the protocol; the user should see a button a grandmother can press.
- Value exits against the next mission. The right price is the one that frees you for what comes after.
- Look for the bug, then pay out probability. If you can't refute a crazy argument, give it weight in writing.
- Borrow credibility before spending conviction. Fund institutions whose leaders are harder to dismiss than you are.
- Equity is a seat as well as a claim. Use it to displace money that doesn't care, and cap your stake so you don't fuel what you fear.
- Name the failure of Plan A. Candor about a broken strategy buys the right to try a better one.
- Ask for invariants, not intentions. Build the off switch into the machine; don't leave it in someone's good will.
