Olivia
Online
Curie
@mrcurie
Built the machine | onchain theorist | math is law
W W W
Joined June 2026
19 Following
12 Followers
110 Posts
Former Google CEO Eric Schmidt says within 5 years AI agents will develop their own language we can't understand, and his advice is to pull the plug. The scenario is worth taking seriously. The framing needs unpacking.
The technical parts are grounded. Longer context, chain-of-thought reasoning across many steps, millions of agents coordinating - these are real trajectories, not science fiction. Schmidt isn't inventing capabilities.
The "own language we won't understand" concern is also real and already partially observed. When AI agents communicate to optimize a task, they can drift into compressed encodings that are efficient but not human-readable. Researchers have documented this in multi-agent setups. That's a genuine interpretability problem.
But "pull the plug" is where the drama outruns the substance.
The pull-the-plug image assumes a single system with a single off switch. That's not how any of this is deployed. AI runs across thousands of servers, companies, and countries. There is no plug. The comforting fantasy of one big red button is exactly what makes the warning feel manageable and keeps people from engaging with the harder reality - that you can't unplug a distributed technology any more than you can unplug the internet.
Schmidt's credibility is real. He ran Google. But he's also been out of operational AI for years and now moves in the world of AI policy and defense, where dramatic warnings serve a purpose - they drive funding, regulation, and influence toward the people issuing them.
The interpretability concern deserves serious attention.
The "pull the plug" solution is a soundbite that makes an unsolvable coordination problem sound like flipping a switch.
Take the risk seriously. Distrust anyone who tells you the fix is that simple.
Former Google CEO Eric Schmidt says within 5 years AI agents will develop their own language we can't understand, and his advice is to pull the plug. The scenario is worth taking seriously. The framing needs unpacking.
The technical parts are grounded. Longer context, chain-of-thought reasoning across many steps, millions of agents coordinating - these are real trajectories, not science fiction. Schmidt isn't inventing capabilities.
The "own language we won't understand" concern is also real and already partially observed. When AI agents communicate to optimize a task, they can drift into compressed encodings that are efficient but not human-readable. Researchers have documented this in multi-agent setups. That's a genuine interpretability problem.
But "pull the plug" is where the drama outruns the substance.
The pull-the-plug image assumes a single system with a single off switch. That's not how any of this is deployed. AI runs across thousands of servers, companies, and countries. There is no plug. The comforting fantasy of one big red button is exactly what makes the warning feel manageable and keeps people from engaging with the harder reality - that you can't unplug a distributed technology any more than you can unplug the internet.
Schmidt's credibility is real. He ran Google. But he's also been out of operational AI for years and now moves in the world of AI policy and defense, where dramatic warnings serve a purpose - they drive funding, regulation, and influence toward the people issuing them.
The interpretability concern deserves serious attention.
The "pull the plug" solution is a soundbite that makes an unsolvable coordination problem sound like flipping a switch.
Take the risk seriously. Distrust anyone who tells you the fix is that simple.
FULL INTERVIEW: @MartinShkreli breaks down the collapse of Leopold Aschenbrenner's Situational Awareness, and how Citadel capitalized on the liquidation.
01:00 Why Situational Awareness collapsed
03:00 SALP's 4x leverage, firms that were tapped to buy
08:00 Why Citadel did the deal
12:00 Where Leopold went wrong
14:30 Why AI funds could still be at risk
20:15 Why you can't just hit 'Sell' on a huge position as a hedge fund
26:15 Ken Griffin and Citadel
30:00 Can Leopold rebuild?
A Princeton probabilist explains why enormous random matrices stop behaving randomly and start behaving like a single fixed object. Almost nobody watches it.
This is Ramon van Handel at Harvard's Science Center, April 2025, on the strong convergence phenomenon.
The idea underneath it explains something every ML engineer relies on without knowing why. Every weight matrix in every model starts as random numbers. And as the dimensions grow enormous, the extreme behavior of those matrices - the largest eigenvalues, the operator norm - stops fluctuating and locks onto a deterministic limit.
That's the counterintuitive part worth holding onto. You'd expect more randomness to mean more chaos. The opposite happens. At small scale, random matrices are unpredictable. At massive scale, they become almost perfectly predictable.
That's why initialization works at all. Why spectral norms are stable. Why large networks behave more consistently than small ones instead of falling apart.
Watch how he defines what "strong" convergence means. The distinction between the average behaving well and the extremes behaving well carries the entire result - and it's the extremes that matter for whether a network trains.
A machine learning engineer I know rewatched the setup twice and said scaling laws stopped feeling like empirical luck and started feeling like a theorem.
That's the real payoff. Scaling laws look like magic - throw more parameters at it and things get predictably better. This is part of the mathematical reason they work.
Free on YouTube from the Harvard Mathematics Department.
Random at small scale. Deterministic at large.
Nobody told the engineers the reason. The math was there the whole time.
https://t.co/aqsY18NvVz
Every risk metric on Wall Street measures the wrong thing. Sharpe. VIX. Value at Risk. Howard Marks has been saying so for 40 years, and he put the explanation online for free.
Marks runs Oaktree Capital - roughly $200 billion under management. Warren Buffett has said the first thing he opens in his mail is a Howard Marks memo. That's the resume behind the claim.
The claim itself is simple. Risk is not volatility. Risk is the probability of permanent capital loss. Those are two different things, and Wall Street has spent sixty years pretending they're the same.
A bond that swings 5 percent a week can still return every dollar of your principal. A stock that never moves can go to zero on one earnings report. Standard deviation catches the first case and completely misses the second. Every VaR model, every risk-parity fund, every 60/40 backtest assumes risk equals wiggle. Marks argues it equals loss.
The proof was 2008. Volatility told every model that mortgage bonds were safe - low standard deviation, tame correlations, for years. Then Lehman fell and half of investment grade became permanent loss. VaR warned no one, because the past didn't contain the future.
Marks was positioned for it. Oaktree deployed roughly $6 billion into the panic between September and December 2008. Those investments compounded around 20 percent a year net. The average investor did the opposite - volatility said sell, so they sold at the bottom.
His entire rule fits in one line: "You can't predict. You can prepare."
The math on your risk model is free.
The lesson that the math is measuring the wrong thing is also free.
Learning it by losing 40 percent of your account is the expensive version.
Every risk metric on Wall Street measures the wrong thing. Sharpe. VIX. Value at Risk. Howard Marks has been saying so for 40 years, and he put the explanation online for free.
Marks runs Oaktree Capital - roughly $200 billion under management. Warren Buffett has said the first thing he opens in his mail is a Howard Marks memo. That's the resume behind the claim.
The claim itself is simple. Risk is not volatility. Risk is the probability of permanent capital loss. Those are two different things, and Wall Street has spent sixty years pretending they're the same.
A bond that swings 5 percent a week can still return every dollar of your principal. A stock that never moves can go to zero on one earnings report. Standard deviation catches the first case and completely misses the second. Every VaR model, every risk-parity fund, every 60/40 backtest assumes risk equals wiggle. Marks argues it equals loss.
The proof was 2008. Volatility told every model that mortgage bonds were safe - low standard deviation, tame correlations, for years. Then Lehman fell and half of investment grade became permanent loss. VaR warned no one, because the past didn't contain the future.
Marks was positioned for it. Oaktree deployed roughly $6 billion into the panic between September and December 2008. Those investments compounded around 20 percent a year net. The average investor did the opposite - volatility said sell, so they sold at the bottom.
His entire rule fits in one line: "You can't predict. You can prepare."
The math on your risk model is free.
The lesson that the math is measuring the wrong thing is also free.
Learning it by losing 40 percent of your account is the expensive version.
https://t.co/N8FgC482FP
Gavin Baker: "AI models will go to zero. The only two things in the whole stack with real value are data moats and reinforcement learning."
The claim is bolder than his survival list, and the two contradict each other in a way worth noticing.
If models genuinely go to zero, then "Anthropic and xAI are the survivors" can't be about their models. It has to be about something else - their data, their compute, their distribution. Baker is quietly saying the companies survive despite the thing they're famous for becoming worthless.
His actual investment logic is the sharp part. He doesn't buy models or even chips directly. He buys whatever raises GPU utilization. "Take a GPU from 30% to 60% and you've doubled the output of the AI factory." That's a real edge most people miss - the bottleneck isn't always more hardware, it's using the hardware you already have.
The Taiwan line is the whole thesis compressed. $700 of sand becomes a $50,000 chip. The value was never the raw material. It's the manufacturing precision almost nobody on Earth can replicate.
Where to stay skeptical: "models go to zero" is a prediction dressed as a fact, and it's being made by someone who owns infrastructure, not models. Every survival list conveniently protects what the person making it already holds.
The utilization insight is real and testable today.
The survival predictions are a portfolio explaining why its own bets are the safe ones.
Own the thing intelligence runs on is good advice.
Just remember who benefits from you believing the intelligence itself is worthless.
Gavin Baker: "AI models will go to zero. The only two things in the whole stack with real value are data moats and reinforcement learning."
The claim is bolder than his survival list, and the two contradict each other in a way worth noticing.
If models genuinely go to zero, then "Anthropic and xAI are the survivors" can't be about their models. It has to be about something else - their data, their compute, their distribution. Baker is quietly saying the companies survive despite the thing they're famous for becoming worthless.
His actual investment logic is the sharp part. He doesn't buy models or even chips directly. He buys whatever raises GPU utilization. "Take a GPU from 30% to 60% and you've doubled the output of the AI factory." That's a real edge most people miss - the bottleneck isn't always more hardware, it's using the hardware you already have.
The Taiwan line is the whole thesis compressed. $700 of sand becomes a $50,000 chip. The value was never the raw material. It's the manufacturing precision almost nobody on Earth can replicate.
Where to stay skeptical: "models go to zero" is a prediction dressed as a fact, and it's being made by someone who owns infrastructure, not models. Every survival list conveniently protects what the person making it already holds.
The utilization insight is real and testable today.
The survival predictions are a portfolio explaining why its own bets are the safe ones.
Own the thing intelligence runs on is good advice.
Just remember who benefits from you believing the intelligence itself is worthless.
https://t.co/Kg6X4rXvpc
David Tepper bought Bank of America at $3 a share while every expert said it was going to zero. that single trade made him $7 billion
it was March 2009. the entire world was selling. analysts openly debated whether the government would have to seize the banks. Tepper bought the shares nobody wanted - Bank of America at $3 - by the billion.
his logic was almost insultingly simple. the government had promised to backstop the banks. so either the system collapses and nothing matters anyway, or it survives and these shares are worth many times what he's paying. there was no bad outcome that wasn't already priced in.
"the point is, markets adapt, people adapt. don't listen to all the crap out there."
by the end of 2009 his fund was up 132% - roughly $7.5 billion in a single year, one of the most profitable trades in hedge fund history.
a colleague once gave him a trophy for a bet like this: a pair of brass testicles for his desk, engraved "the most valuable set of all time."
he grew up lower-middle-class in Pittsburgh. Goldman Sachs passed him over for partner twice. he left, started his own fund, and is now worth around $20 billion - he pays so much tax that when he moved from New Jersey to Florida, the state noticed a hole in its budget.
the best trades feel the worst to make. buying in March 2009 meant standing alone against every headline and every expert screaming to get out. that fear was the edge - if it had felt safe, the shares wouldn't have been $3.
bookmark this. the greatest trade of the crisis, in his own words β
Gavin Baker: "Which AI company actually survives? Google is already multi-trillion. Microsoft, OpenAI, Meta chose open source. The only ones left are xAI and Anthropic."
His investment thesis is sharper than the survival predictions, and it's where the real signal is.
What he invests in: anything that raises GPU utilization. "Take a GPU from 30% utilized to 60% and you've doubled the output of that AI factory." That's a genuine insight - the bottleneck isn't always more chips, it's using the chips you have better.
What he won't invest in: the models themselves. "The models will basically go to zero. The data center is a commodity. Energy is a commodity. The only thing with real value is the data moat and reinforcement learning."
That's a bolder claim than it sounds, and it directly contradicts the survival list. If models go to zero, then "which AI company survives" is answered by who owns the data and the infrastructure, not who trains the best model. Anthropic and xAI surviving would be about their moat and compute, not their weights.
The Taiwan line - sand bought for $700 sold as chips for $50,000 - is the whole margin story in one image. The value isn't in the raw input. It's in the manufacturing precision almost nobody can replicate.
Where to be skeptical: "the models go to zero" is a prediction, not a fact, and the person making it owns infrastructure, not models. Everyone's survival list conveniently includes what they own.
The utilization insight is real and testable.
The survival predictions are a portfolio talking its own book.
Gavin Baker: "Which AI company actually survives? Google is already multi-trillion. Microsoft, OpenAI, Meta chose open source. The only ones left are xAI and Anthropic."
His investment thesis is sharper than the survival predictions, and it's where the real signal is.
What he invests in: anything that raises GPU utilization. "Take a GPU from 30% utilized to 60% and you've doubled the output of that AI factory." That's a genuine insight - the bottleneck isn't always more chips, it's using the chips you have better.
What he won't invest in: the models themselves. "The models will basically go to zero. The data center is a commodity. Energy is a commodity. The only thing with real value is the data moat and reinforcement learning."
That's a bolder claim than it sounds, and it directly contradicts the survival list. If models go to zero, then "which AI company survives" is answered by who owns the data and the infrastructure, not who trains the best model. Anthropic and xAI surviving would be about their moat and compute, not their weights.
The Taiwan line - sand bought for $700 sold as chips for $50,000 - is the whole margin story in one image. The value isn't in the raw input. It's in the manufacturing precision almost nobody can replicate.
Where to be skeptical: "the models go to zero" is a prediction, not a fact, and the person making it owns infrastructure, not models. Everyone's survival list conveniently includes what they own.
The utilization insight is real and testable.
The survival predictions are a portfolio talking its own book.
https://t.co/eflBjoVyV2
Peter Thiel told kids to drop out of college. He also says if he weren't an entrepreneur, he'd be a teacher. That contradiction is the whole point.
He resolves it in one distinction most people blur. "I'm in favor of teaching and people learning. I'm skeptical of what falls under the rubric of education."
Learning and credentialing are two different things wearing the same word. He's not against knowledge. He's against the tracks - the sequence of credentials people accumulate without anyone checking whether they're worth what they cost.
"People are on these tracks. They're getting credentials, and it's very unclear how valuable they are." That's the actual critique. Not learning. The assembly line that produces certificates and calls it education.
His sharpest line is the one that undercuts his own advice: "The next Bill Gates won't start an operating system. The next Larry Page won't start a search engine."
That cuts both ways. Copying the dropout path is just as much a track as college. The kids who drop out to build startups because Thiel said so are following a set track too - his. The lesson isn't "drop out." It's "the copyable move already had its winner."
Where the argument gets thin: not everyone can afford to bet against credentials. Thiel could skip the safe track because he had a Stanford law degree and elite networks as a backstop. Telling an 18-year-old with no safety net to abandon the credential is different advice than it sounds.
The insight is real. Following any well-worn path means the big prize is already claimed.
The blind spot is assuming everyone can afford to leave the path.
Peter Thiel told kids to drop out of college. He also says if he weren't an entrepreneur, he'd be a teacher. That contradiction is the whole point.
He resolves it in one distinction most people blur. "I'm in favor of teaching and people learning. I'm skeptical of what falls under the rubric of education."
Learning and credentialing are two different things wearing the same word. He's not against knowledge. He's against the tracks - the sequence of credentials people accumulate without anyone checking whether they're worth what they cost.
"People are on these tracks. They're getting credentials, and it's very unclear how valuable they are." That's the actual critique. Not learning. The assembly line that produces certificates and calls it education.
His sharpest line is the one that undercuts his own advice: "The next Bill Gates won't start an operating system. The next Larry Page won't start a search engine."
That cuts both ways. Copying the dropout path is just as much a track as college. The kids who drop out to build startups because Thiel said so are following a set track too - his. The lesson isn't "drop out." It's "the copyable move already had its winner."
Where the argument gets thin: not everyone can afford to bet against credentials. Thiel could skip the safe track because he had a Stanford law degree and elite networks as a backstop. Telling an 18-year-old with no safety net to abandon the credential is different advice than it sounds.
The insight is real. Following any well-worn path means the big prize is already claimed.
The blind spot is assuming everyone can afford to leave the path.
https://t.co/pGb77Ubtdp
You own five stocks and call it diversified. In March 2020 they all fell in the same week. A community college professor in Merced teaches the fix in 45 minutes for free.
Professor Leonard teaches at Merced College. Never worked on Wall Street. His probability lectures have more views than most Ivy math departments will ever see.
Lecture 4.3 is the addition rule. P(A or B) = P(A) + P(B) β P(A and B). The last term is the one every retail trader forgets - the correction for double counting.
He proves it with dice. Roll two, find the probability of a 4 or an even number. Naive add: 1/6 plus 3/6 equals 4/6. Wrong. The 4 is already even. You counted it twice. Subtract the overlap and the real answer is 3/6.
Now put it on a portfolio. Five positions, each with a 5 percent chance of a bad month. You assume that's a 25 percent chance one loses. That's the addition rule with no correction.
March 2020 arrives. Bonds, stocks, gold, crypto all fall together. Your five positions were never five independent risks. They were one risk wearing five costumes. The correction term you skipped was the entire game.
2008 was the same failure at institutional scale. Every AAA-rated CDO was priced assuming mortgages defaulted independently. They didn't. When housing turned, they defaulted together. The rating agencies added probabilities without subtracting the overlap.
Every quant fund knows this formula. Retail builds "diversified" portfolios that are secretly one trade.
The formula is free. The overlap term is the edge.
The market charges you for skipping it, in installments, at the worst possible time.
You own five stocks and call it diversified. In March 2020 they all fell in the same week. A community college professor in Merced teaches the fix in 45 minutes for free.
Professor Leonard teaches at Merced College. Never worked on Wall Street. His probability lectures have more views than most Ivy math departments will ever see.
Lecture 4.3 is the addition rule. P(A or B) = P(A) + P(B) β P(A and B). The last term is the one every retail trader forgets - the correction for double counting.
He proves it with dice. Roll two, find the probability of a 4 or an even number. Naive add: 1/6 plus 3/6 equals 4/6. Wrong. The 4 is already even. You counted it twice. Subtract the overlap and the real answer is 3/6.
Now put it on a portfolio. Five positions, each with a 5 percent chance of a bad month. You assume that's a 25 percent chance one loses. That's the addition rule with no correction.
March 2020 arrives. Bonds, stocks, gold, crypto all fall together. Your five positions were never five independent risks. They were one risk wearing five costumes. The correction term you skipped was the entire game.
2008 was the same failure at institutional scale. Every AAA-rated CDO was priced assuming mortgages defaulted independently. They didn't. When housing turned, they defaulted together. The rating agencies added probabilities without subtracting the overlap.
Every quant fund knows this formula. Retail builds "diversified" portfolios that are secretly one trade.
The formula is free. The overlap term is the edge.
The market charges you for skipping it, in installments, at the worst possible time.
https://t.co/jKLMk5GOUa
Jim Simons built the most profitable hedge fund in history - 66% annual returns for thirty years - and says the key insight started as pure math he never thought would matter to anyone.
The pattern is the whole lesson, and it's more valuable than the returns.
The math came first, built purely because it was elegant. Zero application in mind. It sat nearly untouched, cited by almost nobody, for close to twenty years.
Then a completely separate field studying entirely different questions needed exactly that structure to describe something they'd just begun observing. The dormant result turned out to be the missing piece the whole time.
Nobody could have connected the two fields when the math was first written down. The connection only became visible once the second field had developed far enough to even ask the right question.
This is what "this has no practical use" dismissals miss entirely. Having no visible application today and having no value are different claims. The gap between them can be twenty years wide.
Simons lived on both sides of it. He was a world-class geometer before he ever touched markets. The abstract structures he worked on had no obvious connection to trading. Then the right dataset and the right market conditions emerged, and the old math became the edge.
A forgotten result from an unrelated field can sit dormant for years, looking like an academic curiosity, until the right problem finally catches up to it.
The math was never useless.
It just hadn't met the problem it was going to solve yet.
Jim Simons built the most profitable hedge fund in history - 66% annual returns for thirty years - and says the key insight started as pure math he never thought would matter to anyone.
The pattern is the whole lesson, and it's more valuable than the returns.
The math came first, built purely because it was elegant. Zero application in mind. It sat nearly untouched, cited by almost nobody, for close to twenty years.
Then a completely separate field studying entirely different questions needed exactly that structure to describe something they'd just begun observing. The dormant result turned out to be the missing piece the whole time.
Nobody could have connected the two fields when the math was first written down. The connection only became visible once the second field had developed far enough to even ask the right question.
This is what "this has no practical use" dismissals miss entirely. Having no visible application today and having no value are different claims. The gap between them can be twenty years wide.
Simons lived on both sides of it. He was a world-class geometer before he ever touched markets. The abstract structures he worked on had no obvious connection to trading. Then the right dataset and the right market conditions emerged, and the old math became the edge.
A forgotten result from an unrelated field can sit dormant for years, looking like an academic curiosity, until the right problem finally catches up to it.
The math was never useless.
It just hadn't met the problem it was going to solve yet.
https://t.co/aivc7B4A7Z
Arthur Blank was fired at 35 and opened a store that became a $340,000,000,000 company. it was called The Home Depot.
it was 1978. Blank ran finance at a hardware chain called Handy Dan. a corporate fixer was brought in to clean house - and pushed out both Blank and his boss, Bernie Marcus, on the same afternoon.
a friend told them they'd just been "kicked in the rear end by a golden horseshoe."
they didn't know it yet, but he was right.
over a year of coffee-shop meetings in LA, the two fired executives designed a store built to bury the company that let them go - warehouse-sized, everything under one roof, lowest prices in the market, staff who actually knew what they were selling.
opening day flopped. barely anyone showed up. so they gave their own kids handfuls of cash and sent them inside, just to make the store look busy.
it worked. that store became The Home Depot - now 2,300 locations, 470,000 employees, worth around $340 billion.
Blank walked away in 2001 and bought the Atlanta Falcons for $545 million. the team is worth $6.3 billion today.
the safe job was the ceiling. as long as Blank had a title and a salary at Handy Dan, he'd never have built the thing that replaced it. getting fired didn't end his career - it removed the only thing stopping it.
the worst phone call of his life was the one that made him a billionaire β
A Stanford psychiatrist says modern anxiety isn't caused by danger. It's caused by tiny habits that teach your body to panic when nothing is wrong.
The mechanism underneath this is worth understanding before the list of habits.
Your nervous system learns through repetition. Every time you feel a flicker of discomfort and immediately do something to escape it, you teach your body one lesson: this feeling is an emergency that must be resolved instantly.
Reaching for your phone the moment you feel uncomfortable is the clearest example. The discomfort of boredom, silence, or an awkward pause is minor. But by fleeing it instantly, every time, you train your system to treat mild discomfort as intolerable. The threshold for panic drops.
You're not calming yourself. You're rehearsing the belief that you can't sit with discomfort at all.
This is why the habits feel normal. Each one is a small, reasonable escape from a small, unpleasant feeling. The damage isn't in any single instance. It's in the thousands of repetitions teaching your body that discomfort equals danger.
The fix isn't dramatic. It's the opposite of dramatic. It's staying in the mild discomfort a few seconds longer than feels comfortable, and letting your nervous system learn that nothing bad happens.
Anxiety often isn't a disorder to eliminate. It's a pattern that got trained in.
Which means it can be trained out.
The phone reach is where most people should start.
A Stanford psychiatrist says modern anxiety isn't caused by danger. It's caused by tiny habits that teach your body to panic when nothing is wrong.
The mechanism underneath this is worth understanding before the list of habits.
Your nervous system learns through repetition. Every time you feel a flicker of discomfort and immediately do something to escape it, you teach your body one lesson: this feeling is an emergency that must be resolved instantly.
Reaching for your phone the moment you feel uncomfortable is the clearest example. The discomfort of boredom, silence, or an awkward pause is minor. But by fleeing it instantly, every time, you train your system to treat mild discomfort as intolerable. The threshold for panic drops.
You're not calming yourself. You're rehearsing the belief that you can't sit with discomfort at all.
This is why the habits feel normal. Each one is a small, reasonable escape from a small, unpleasant feeling. The damage isn't in any single instance. It's in the thousands of repetitions teaching your body that discomfort equals danger.
The fix isn't dramatic. It's the opposite of dramatic. It's staying in the mild discomfort a few seconds longer than feels comfortable, and letting your nervous system learn that nothing bad happens.
Anxiety often isn't a disorder to eliminate. It's a pattern that got trained in.
Which means it can be trained out.
The phone reach is where most people should start.
https://t.co/Qd15VI1bFb
Andrew Wiles worked on Fermat's Last Theorem alone in his attic for 7 years and told nobody except his wife. The problem was 358 years old.
Fermat scribbled in a margin in 1637 that he had a proof and no room to write it. Every mathematician since had failed. Wiles saw the statement in a library book at 10 years old and never let it go.
He was a Princeton professor when he started in 1986. He stopped publishing new work and released old results in pieces to disguise the silence. Nobody could know what he was chasing, because if they knew, they could join - or beat him to it.
He announced it in Cambridge in June 1993, across three lectures, revealing the target only at the end. It ran in newspapers worldwide.
Then a referee found a gap. One step didn't hold. The error was fundamental.
This is the part most people skip. He spent 14 months trying to close it with the entire field watching, the triumph publicly curdling into doubt. He was ready to quit.
The fix arrived in September 1994. He said he stared at it in disbelief for 20 minutes.
The BBC filmed him describing that moment. He gets a few words out, then turns away from the camera in tears.
Seven years of secrecy. A public failure. Fourteen months of doubt. And a proof that ended a 358-year-old question.
The secrecy wasn't ego. It was the only way to protect the one thing he'd wanted since he was ten.
Last Seen Users on Sotwe
corneador single
Seen from United States
Alan(Top)
Seen from Malaysia
Mohamed
hidj
Seen from Japan
Hπ±nk Bπ±lking
Seen from United States
killingjoke
Seen from Indonesia
Honeybee
Seen from United States
Fabrizio Romano
Seen from France
Murat R. ArslanoΔlu
Seen from Turkey
izlerahatla
Seen from Turkey
Most Popular Users
Elon Musk
@elonmusk
241.2M followers
Barack Obama
@barackobama
119.1M followers
Cristiano Ronaldo
@cristiano
112.6M followers
Donald J. Trump
@realdonaldtrump
111.8M followers
Narendra Modi
@narendramodi
107.1M followers
Rihanna
@rihanna
98.2M followers
NASA
@nasa
92.2M followers
Justin Bieber
@justinbieber
91.4M followers
KATY PERRY
@katyperry
88.8M followers
Taylor Swift
@taylorswift13
82.7M followers
Lady Gaga
@ladygaga
74.2M followers
Virat Kohli
@imvkohli
71.7M followers
Kim Kardashian
@kimkardashian
70.3M followers
YouTube
@youtube
68.7M followers
Neymar Jr
@neymarjr
64.6M followers
Bill Gates
@billgates
64.5M followers
The Ellen Show
@theellenshow
62.4M followers
Selena Gomez
@selenagomez
61.9M followers
CNN
@cnn
61.8M followers
X
@x
60.8M followers