Nobel Prize winner joins Anthropic! Today, AlphaFold core leader John Jumper officially announced that he will leave Google DeepMind, where he has worked for nearly 9 years, to join Anthropic. The Nobel Prize winner who rewrote the entire structural biology with an AI model turned around and left.
Hassabis quickly responded: "Thank you John for your extraordinary partnership over the past 9 years! What we have achieved with AlphaFold has changed the world."
After 9 years of cooperation and sharing the Nobel Prize, this is probably the most honorable farewell in the science and technology circle.
Just two days ago, Noam Shazeer, the legendary co-leader of the Transformer paper and co-head of Gemini, just announced that he was leaving Google and went to OpenAI.
In less than 72 hours, Google lost two trump cards in a row.
I didn’t keep the one I bought for 2.7 billion U.S. dollars, and I didn’t keep the one I had been in love with for 9 years.
6 months after graduating with Ph.D., directly leading AlphaFold team
In the world of life sciences, John Jumper can be said to be synonymous with “rewriting the entire discipline with AI.”
Jumper was born in Little Rock, Arkansas, in 1985, an ordinary small town in the South of the United States.
I received a double major in mathematics and physics from Vanderbilt as an undergraduate, and then entered the University of Chicago to study for a Ph.D. in theoretical chemistry. Specifically, computational methods are used to simulate the dynamic behavior of proteins.
Mathematics gave him the intuition for modeling, physics gave him an understanding of complex systems, and theoretical chemistry allowed him to understand the protein problem itself better than any pure AI researcher.
The three directions taken together happen to be the scarcest combination of knowledge to solve the protein folding problem.
After receiving his PhD in 2017, Jumper joined DeepMind directly.
It is worth noting that he had almost no deep learning experience at that time. The most prominent thing on his resume was not his mastery of neural networks, but his understanding of protein physics.
But this is exactly what Hassabis likes.
Immediately afterwards, he made a decision that no one expected - let this young man, who had graduated only 6 months ago and had to learn deep learning by doing, directly lead the AlphaFold team.
There is no transition period, no "work as a researcher for a few years to gain qualifications."
Hassabis is betting that to solve the problem of protein folding, understanding proteins is more important than understanding AI. What Jumper took was the biggest gamble in the entire field of computational biology.
One person doubled biology 1000 times
What happened in the next few years can only be described as "outrageous"——
In 2018, AlphaFold made its debut at the protein structure prediction competition CASP, crushing traditional methods.
In 2020, AlphaFold 2 was born, and the protein folding problem that had troubled biologists for 50 years was directly "solved" by an AI model.
In 2021, Jumper led a team to calculate the 3D structure of almost all more than 50,000 human proteins. In the end, about 1 million species and nearly 200 million known protein structures were generated.
Before AlphaFold, humans spent decades using X-ray crystallography, cryo-electron microscopy and other experimental methods to solve a total of approximately 200,000 protein structures.
Jumper’s team doubled its value 1,000 times in one go.
It is no exaggeration to say that AlphaFold has completed the work that biologists have not completed in the past hundred years in a few months.
In May 2024, AlphaFold 3 will be released - not only predicting proteins, but also the interactions between DNA, RNA, and small molecule drugs. The protein-ligand docking accuracy is 76.4%, which is 1.8 times higher than the previous generation method.
Five months later in Stockholm, John Jumper and Demis Hassabis stood on the podium together to receive the Nobel Prize in Chemistry.
Jumper was 39 years old that year and was the youngest Nobel Prize winner in Chemistry in 70 years.
It only took him 7 years from a PhD graduate who had to study deep learning to stand in the spotlight in Stockholm.
At this point, the rate of return on Hassabis's bet is probably one of the best in the history of human science.
So today he passed away. The pain of Google DeepMind is not just as simple as losing a Director.
What happened to Google?
After the news exploded, the comment area on X went into overdrive.
Netizen Chubby said: "This is a huge loss for Google, and it is crazy for Anthropic!"
Some netizens lamented that "Anthropic has welcomed a Nobel Prize winner, and talents are continuing to concentrate on OpenAI and Anthropic." Others shouted directly: "First it was Karpathy, and now it is the people behind AlphaFold. Anthropic is forming the AI Avengers."
Logan Kilpatrick joked that he expected Jumper to "win another Nobel Prize." The tone is teasing, but if you think about it carefully, it's really not an exaggeration.
After the shock, everyone was asking the same question—what happened to Google?
Jumper didn't say it, Anthropic didn't say it, and Google didn't say it either.
Perhaps, a comment from investor Lior Alexander is the closest to the answer so far——
"The cutting-edge AI lab is selling something that Google can't give: the feeling that one person can change the trajectory of the company."
Even those who bought it for 2.7 billion US dollars did not retain it.
Just two days before Jumper's official announcement, Noam Shazeer announced that he was leaving Google and joining OpenAI as "architectural research leader."
He is one of the core authors of the 2017 article "Attention Is All You Need", which was the foundation of modern AI. He designed the multi-headed attention, and he typed out the first usable implementation that outperformed SOTA line by line.
Google spent $2.7 billion to bring him back from Character.AI.
After returning, Shazeer served as the co-head of Gemini and became the number one contributor to Google's large model counterattack.
As a result, less than two years later, he left again. Two days later, Jumper also left.
They are neither the first nor the last.
In the past eight years, more than 20 top researchers signed on landmark papers have left DeepMind/Brain.
In 2025 alone, at least 11 senior executives will leave. DeepMind co-founder Mustafa Suleyman himself was poached by Microsoft in a $650 million acquisition round.
Life sciences, the next battlefield for the three AI giants
Back to Anthropic. The layout started more than two months ago.
On April 3, Anthropic acquired biotech company Coefficient Bio for $400 million in stock. The team has less than 10 people, but it has already made top results in the industry in the field of AI-driven antibody design.
At the same time, Anthropic is also building its own wet laboratory. In October last year, it launched Claude for Life Sciences to help researchers accelerate drug discovery and biological experiment design. In January this year, it launched Claude for Healthcare for medical institutions.
They said the goal is to compress the life sciences R&D cycle by 10 times. And now, a Nobel Prize-level protein scientist is here to take charge of this matter.
In fact, Anthropic isn't the only one betting on life sciences.
OpenAI released GPT-Rosalind, an inference model specifically for biomedicine in April this year. It focuses on drug discovery, genome analysis and protein engineering. It has already reached cooperation with leading pharmaceutical companies such as Amgen, Moderna, and Thermo Fisher.
The OpenAI Foundation directly stated that it will invest no less than US$1 billion in life sciences in the next year. Coupled with the newly recruited Shazeer who is in charge of architecture research, OpenAI is also gaining momentum on this track.
As for Google DeepMind, Isomorphic Labs, a subsidiary of Hassabis, raised US$600 million last year and signed cooperation agreements with Eli Lilly and Novartis with a total milestone value of up to US$3 billion. AlphaFold's technology base is still the industry benchmark.
The three laboratories are pushing their chips in the same direction at the same time - using AI to rewrite life sciences.
Jumper's choice is just the latest step in this big game.