Top-tier AI talents dare not get married
In a tech podcast episode of No Priors, Elad Gil, a well-known Silicon Valley investor, casually mentioned an incident that made people's hearts sink.
He said he knows researchers at several top AI labs, who have recently been seriously discussing a question — "Should I get married?" It's not because their relationships are having problems, but because they believe that:
In 18 months, the world may be a completely different place.
AI may achieve recursive self-improvement, at which point human researchers will no longer be needed. Given the uncertainty of the future, does marriage still make sense?
After listening, Sarah Guo, Elad's partner and fellow investor, said one word — tragic. She compared this mindset to the state of "people feeling like they are about to die".
It is not physical death, but the death of their professional life as a valuable intellectual worker.
This is not a joke. It is the mental state that a group of the smartest people on the planet are going through in 2026.
01
Panicked AI Talents
Over the past six months, the AI industry has witnessed a rare wave of "departures".
Different from previous job hopping, these people left not with better offers, but with open letters written in heavy tones.
On February 9, 2026, Mrinank Sharma, head of the safety research team at Anthropic, posted his resignation letter on X, writing that "the world is in danger". The letter got 14.7 million views.
Two days later, Zoë Hitzig, a researcher at OpenAI, published an article in *The New York Times* announcing her resignation. Going further back, Jan Leike, former head of OpenAI's Superalignment team, and Ilya Sutskever, co-founder of OpenAI, also left one after another — the latter founded Safe Superintelligence Inc., raised 3 billion US dollars, and has not released any products to date.
These people are not too exhausted to move on. On the contrary, they are more clear-headed than anyone else when they leave.
The resignation post from Hieu Pham, a former employee of OpenAI and xAI, is probably the most candid one. This researcher who helped build the world's most powerful AI system wrote: "I used to scoff at the idea of deteriorating mental health. But it is real, painful, terrifying and dangerous." He returned to Vietnam with his family, saying he wanted to "find a way to heal".
Nathan Lambert, senior research scientist at AI2, described the norm inside the labs on Lex Fridman's podcast — the work intensity at OpenAI and Anthropic is already equivalent to the "996" culture.
Back to that episode of the No Priors podcast, Elad Gil presented the logical chain behind this crazy pace.
The mainstream belief in the labs goes like this — programming problems will be basically solved by the end of 2026, and some form of mild recursive self-improvement (RSI) will appear by the end of 2027, at which point models will start training most of the models themselves. "If my career only has a year and a half of effective time left," one researcher calculated, "then each week accounts for 2% of my remaining productivity. So I have to work 16 hours a day."
This set of calculations is chillingly precise. But Sarah Guo raised a key counter-question — in the past five years, the prophecy that "ASI will arrive in 18 months" has been repeated every 18 months. So how reliable is it as a prediction tool?
No one can answer this question. But that doesn't stop top AI talents from arranging their lives according to it.
02
Writing Eulogies for the Mathematicians' Career
If the anxiety inside AI labs can still be interpreted as an overreaction of "being at the center of the vortex", the same emotion is spreading to a seemingly more distant group — pure mathematicians.
2026 is known as a turning point for the mathematics field. LLM is proving professional mathematician-level theorems almost every week. OpenAI's internal models release batches of mathematical breakthroughs at once. A sentence is circulating on X — "This year's Fields Prize will be the last one."
The Fields Prize was indeed awarded to four outstanding mathematicians in July 2026. But the headlines of media reports reveal an ominous undertone — AFP used the wording "as AI reshapes the mathematics industry". Stanford University specially held a seminar, inviting three Fields Prize winners and researchers from OpenAI and DeepMind, with the topic of "the future of mathematics". Terence Tao proposed at the meeting that people need to go beyond the question of "whether AI can generate proofs" and rethink the mathematical workflow itself.
A discipline that has existed for thousands of years suddenly has to answer the question "do we still need to exist" within a year.
*Boston Review* published a long article in the summer of 2026, titled "The Collapse of Knowledge". The article points out that from *The New York Times* to *Science* and *Nature*, the public has been repeatedly told that mathematics will be the next domino to fall on AI's "road to AGI". A syllogism is circulating in the mathematician circle with growing urgency — AI can do mathematics → AI is improving rapidly → so AI will "solve" mathematics.
One researcher wrote an article titled "Existential Risks from AI — A Note for Mathematicians", starting with: "2026 is a critical year for mathematics. LLMs prove professional-level theorems every week with minimal guidance." He wrote that this story could be a dark science fiction novel — "We fight each other for the Fields Prize until the bitter end, never really seeing it coming".
This emotion does not only exist in marginal voices. Fields Prize winner Mike Freedman is now the chief scientific officer of Logical Intelligence, an AI company co-founded by Yan LeCun. Fields Prize winner Tim Gowers publicly admitted that AI has solved important conjectures that he "has heard of but are not in his own field". When the top people in a discipline start to admit in a complex tone that machines are catching up, it is not a bluff.
03
The Weird "AI Psychosis"
When enough people show similar symptoms, the psychiatric community will give it a name.
In September 2025, two psychiatric researchers Stephanie McNamara and Joseph Thornton from the University of Florida College of Medicine published a paper in the academic journal Cureus, officially proposing a new clinical concept — "Artificial Intelligence Replacement Dysfunction, AIRD".
AIRD is not ordinary unemployment anxiety. It attacks the most core part of a person — "who I am".
Thornton calls it an "invisible disaster". Symptoms of AIRD include anxiety, insomnia, paranoia, loss of identity and sense of worthlessness, and can appear in people without any previous history of mental illness. The paper points out that this pain "is not rooted in traditional psychopathology, but in the existential threat of professional obsolescence".
There is a key distinction here. The pain of textile workers, carriage drivers and typists eliminated by previous technological waves was mainly economic, as they lost their source of income and needed to support their families. The pain of AIRD patients is existential — "If AI can do everything I do, what do my 20 years of training and accumulation mean? What unique value does my mind still have?"
A study from the University of Mannheim specifically examined doctors' reactions to AI, and found that the perception of "threat to professional competence" and "threat to professional recognition" directly leads to a resistant attitude towards AI. Interestingly, medical students show a stronger sense of identity threat and resistance than experienced doctors — young people who have not had time to fully establish their professional identity are more easily shaken by AI.
A KPMG survey shows that 52% of employees worry that AI will threaten their job security. And Quantum Workplace's analysis of the voices of more than 700,000 employees found that employees who use AI frequently report a burnout rate of 45%, much higher than the 35% of people who do not use AI. The more they use it, the more anxious they become — this is a data point worth pondering in itself.
04
The Fourth "Decentralization" Shock
Why is the impact of AI technology different from any previous technological revolution?
Some researchers have proposed a highly impactful framework. Human history has experienced three fundamental "decentralization" shocks:
Copernicus told humanity that you are not the center of the universe;
Darwin told humanity that you are not the center of the species;
Freud told humanity that you are not even the master of your own inner mind.
AI is the fourth shock. It tells humanity — your cognitive ability and intelligence are not unique either.
All previous technological revolutions had an unchanging safe haven. The Industrial Revolution amplified human muscle power, and the Internet amplified human communication capabilities, but they were all doing the same thing — making the human mind more valuable. The loom replaced weavers, but made textile engineers more valuable. Cars wiped out the carriage industry, but made mechanical engineers the darlings of the era.
The more advanced technology becomes, the higher the premium for "mental work". Top talents are not only not anxious, but also the biggest beneficiaries of every wave of technological progress.
Some people call what is happening now the "Great Reversal" — for the first time in human history, the most elite cognitive tasks are the most vulnerable to cutting-edge technology. Programming, mathematical proof, legal analysis, scientific research... These jobs that were once considered to require years of training and high intelligence to do well are precisely the fields where AI is making the fastest progress.
Previous technological revolutions changed people's way of working, but never shook the status of work as a "source of meaning". This time, even the meaning itself is being shaken.
What is even more anxiety-provoking is the speed of change.
The Industrial Revolution unfolded over several generations, leaving a window for society to adapt and transform. AI iterates on a monthly basis. Elad Gil said in the podcast, "One year in AI time is equivalent to three to four years in the normal cycle." Systems that struggled on certain tasks a few months ago can crush entry-level white-collar workers a few months later.
People have no time to redefine themselves before the world has already changed.
Back to Sarah Guo's sharp observation — in the past five years, the prophecy that "recursive self-improvement will arrive in 18 months" has been repeated every 18 months by a group of smart, even self-aware researchers. It has not come true yet.
But the way it is coming true may be more subtle than people realize.
Not through a sudden technological leap, but through continuous erosion of human psychology. Even if ASI does not arrive tomorrow, even if RSI takes five years instead of 18 months, this belief itself has already changed people's behavior. Researchers dare not get married, mathematicians start writing eulogies for their own careers, and psychiatrists have to give a name to this new type of anxiety.
This is the most noteworthy part of this "doomsday mindset". It may be a rational technical prediction, it may be a collective hysteria of the tech elite circle, or more likely, it is both.
This is a continuously self-reinforcing anxiety spiral built on real technological progress.
Hieu Pham returned to Vietnam to find healing, Ilya Sutskever raised 3 billion US dollars to build "safe superintelligence". Mrinank Sharma's resignation letter was read by 14.7 million people. These choices are completely different, but the question behind them is the same — when you are personally building something that may make you obsolete, how should you live?
Maybe Sarah Guo is right. The answer is simpler than it sounds — you should get married. You should continue to live your life.
Not because the prophecy will not come true, but because whether it comes true or not, it is something worth doing.
This article is from the WeChat Official Account "GeekPark" (ID: geekpark), author: Yuhang Yuan, editor: Jing Yu, published with authorization from 36Kr.