Original editor: Marco
Original source: New Intelligence
On May 19th at 11 PM, Andrej Karpathy officially announced his joining Anthropic.
The weight of this name needs no further explanation.
Co-founder of OpenAI, former AI director of Tesla, father of "Vibe Coding," the most influential AI educator in the world.
His status in the AI field is roughly equivalent to that of LeBron James in basketball; he is headline news wherever he goes.
He only posted three sentences on X.
https://x.com/karpathy/status/2056753169888334312
The first sentence says that the future years of LLM frontiers are "particularly formative." The third sentence states that he still loves education. The most crucial middle sentence contains five words: "Return to research and development."
This is the third core figure from OpenAI to join Anthropic within two years.
He is also someone who is about to turn 40, has achieved success, and is financially independent, choosing to work under someone else.
Why leave? Why Anthropic? Why does Anthropic want him so badly?
Every question has a layer worth unpacking.
What he will do
Karpathy has already started working this week, joining the pre-training team at Anthropic.
This team is led by Nick Joseph and is responsible for all large-scale training operations for Claude.
An Anthropic spokesperson confirmed to TechCrunch that Karpathy will form a new sub-team focusing on accelerating pre-training research using Claude itself.
Nick Joseph also added background on X, saying, "He will build a team focused on accelerating pre-training research with Claude itself."
https://x.com/nickevanjoseph/status/2056760504949842219
TechCrunch commented, "Karpathy is one of the few researchers capable of bridging the gap between LLM theory and large-scale training practice."
Axios characterized this as "a significant victory for Anthropic in the talent war."
On the same day, cybersecurity expert Chris Rohlf also announced joining Anthropic, and earlier this month, former xAI founding member Ross Nordeen joined as well. The direction of talent flow is increasingly clear.
https://x.com/chrisrohlf/status/2056744653165092983
Data from Polymarket can serve as a secondary proof of market sentiment—traders have priced the probability of Anthropic having the best AI model by the end of June at 65%, while OpenAI is at 4%.
https://polymarket.com/event/which-company-has-best-ai-model-end-of-june
Karpathy's joining further solidifies this judgment.
The definer Karpathy
To understand the weight of this joining, one must understand the rarity of Karpathy as a person.
His rarity does not lie in his technical abilities; there are many top researchers.
His rarity lies in his ability to change the way an entire industry understands something with a single word.
Born in 1986 in Slovakia, he immigrated to Toronto, Canada at the age of 15.
While studying at the University of Toronto, he took a course from Geoffrey Hinton and participated in his reading group.
Hinton is the spiritual leader of the deep learning renaissance, Turing Award winner in 2018, and Nobel Prize laureate in Physics in 2024.
Karpathy was one of the young individuals ignited by this fire early on.
He then studied under another legendary figure, Fei-Fei Li at Stanford, where he created the CS231n course during his PhD.
This course grew from 150 students in 2015 to 750 in 2017, with all video lectures publicly available online, becoming the first stop for countless engineers learning deep learning, and is considered the top course in computer vision, without exception.
In 2015, he became a founding research scientist at OpenAI.
In 2017, he was recruited by Musk to Tesla as a senior director of AI, pushing autonomous driving towards a purely visual solution.
Musk faced significant pressure during this recruitment.
https://www.cnbc.com/2026/05/19/anthropic-hires-openai-cofounder-andrej-karpathy-former-tesla-ai-lead.html
That same year, Karpathy published an article on Medium proposing the concept of "Software 2.0," advocating that neural network weights are the new code, datasets are the new source code, and gradient descent is the new compiler.
This framework reshaped the entire industry's understanding of "what programming is."
After leaving Tesla in 2022, he created the "Neural Networks: Zero to Hero" series on YouTube, gaining over a million subscribers.
The concurrent open-source projects micrograd, nanoGPT, and nanochat, with minimal code but precisely hitting core concepts, have been referred to as "executable textbooks."
In February 2025, he coined the term "Vibe Coding," which was selected as word of the year by Collins Dictionary.
https://x.com/karpathy/status/1886192184808149383
In June, during a speech at YC AI Startup School, he introduced the frameworks "Software 3.0" and "The Decade of Agents," becoming one of the most widely discussed AI talks that year.
TIME listed him as one of the "100 Most Influential People in AI" in 2024.
From Hinton to Li Fei-Fei to Altman and then to Musk, he has been at the forefront at every node.
However, what he left behind that is most enduring is not any product or paper, but those conceptual frameworks.
Software 2.0, Vibe Coding, LLM OS. These terms changed the way people think about AI.
Why willingly take a "minus two"
Karpathy's career has a clear thread; he has never chased titles.
He has been a student of Hinton and Li Fei-Fei, a colleague of Altman, and a direct subordinate of Musk.
In every experience, his organizational position has been high-level.
Now he joins Anthropic, his direct supervisor is Nick Joseph, head of pre-training.
Nick Joseph reports to Dario Amodei.
Karpathy is positioned at the third level in the organizational structure.
Nick Joseph is one of the 11 founding members of Anthropic, having previously worked at Vicarious and OpenAI.
During his time at OpenAI, he worked on code models in the safety team, observing that GPT-3 could write code after fine-tuning, realizing that AI could self-improve, and subsequently left with the safety team leader to create Anthropic.
His team trained the entire series of Claude models, including Mythos.
Karpathy is willing to conduct research under Nick Joseph for a simple reason: this position is closest to what he wants to do.
Tracing back through each of his career moves, the driving force has always been the same: "Where is the biggest experiment at this moment."
In 2017, he went to Tesla because autonomous driving was the biggest experimental ground for Software 2.0.
In 2022, he left because the architecture was set, and what remained was engineering optimization.
In 2023, he returned to OpenAI because the explosive period following the release of GPT-4 with ChatGPT was the most stimulating frontier.
In 2024, he founded Eureka Labs to test the hypothesis of AI-native education.
In 2026, he joined Anthropic because the pre-training revolution of "using AI to research AI" is happening here.
Every departure has not been out of dissatisfaction, but rather because the current position is no longer where the biggest experiment is located.
Why not return to OpenAI? The direction of talent flow provides the answer.
Jan Leike, former alignment lead at OpenAI, joined Anthropic in May 2024.
OpenAI co-founder John Schulman followed suit in August of the same year.
Now it is Karpathy's turn.
In two years, three people have all moved in one direction, with no comparable reverse cases.
OpenAI's strategic focus has shifted from pure research to platformization and acquisitions. Chat.com, io Products, Windsurf, TBPN, the intervals between acquisitions are getting shorter, and the amounts larger.
This is a company that is turning into a "consumer giant of the AI era."
For a researcher looking to "return to R&D," Anthropic's route of "winning by research quality" is more attractive.
Why Anthropic wants him so much
Anthropic's hiring motivations can be divided into several layers.
The most surface layer is technical need.
No matter how large Anthropic's compute budget is, it cannot compare to the backing of Microsoft for OpenAI and Google with its TPU.
In a pure compute power race, Anthropic cannot win.
It must find a way to train better models with less compute power.
"Using Claude to accelerate pre-training research" is this route, and Karpathy possesses a rare combination of depth in pre-training theory, large-scale engineering experience, and intuition for AI-assisted research.
Going further down is talent signal.
In two years, three core figures from OpenAI have moved unidirectionally to Anthropic, and the narrative of "first-line researchers voting with their feet" has taken shape.
Every new addition of Karpathy's caliber lowers the psychological barrier for the next top talent to join. Talent attracts talent, creating a self-reinforcing cycle.
Additionally, there is the brand enhancement before an IPO.
Anthropic is negotiating a $30 billion financing at a $900 billion valuation, and preparations for an IPO are underway.
Karpathy is one of the most publicly recognized technical figures in the AI field, with a million YouTube subscribers, coiner of a word of the year, and the CLAUDE.md repository with 220,000 GitHub stars.
His name appearing on Anthropic's employee list directly provides an investment bank with a line to write into the prospectus.
However, the most interesting layer that Anthropic might not explicitly consider as a hiring motivation, but will yield the greatest returns, is Karpathy's ability to define paradigms.
Any technical exploration he undertakes at Anthropic will be openly discussed by him in tweets, blogs, and YouTube videos.
When he names what is happening in his unique way, Anthropic naturally becomes the birthplace of that paradigm.
By hiring a top pre-training researcher, they also gain the industry's most influential technical storyteller.
The critical point of the flywheel
Viewing this personnel change in a larger context marks a technological inflection point.
In April 2026, Anthropic released Mythos Preview, the most powerful AI model to date.
Mythos is too powerful and can only be invited for internal testing through Project Glasswing
Without specialized training in cybersecurity, Mythos autonomously discovered and exploited a 17-year-old remote code execution vulnerability in FreeBSD, found a 27-year-old vulnerability in OpenBSD, and a 16-year defect in FFmpeg.
Independent evaluation from the UK's AI safety research institute confirmed it is the first model capable of completing a 32-step enterprise network attack simulation from start to finish.
Anthropic also admits that these capabilities are not the result of deliberate training, but "downstream emergence" from enhancements in general reasoning and software engineering capabilities.
The better the pre-training is done, the more the emergent capabilities exceed expectations.
Mythos is currently the most powerful model and the most powerful tool.
What Karpathy is going to do at Anthropic is to take this strongest hammer and improve the way the hammer itself is made.
Using Mythos / Claude to discover better training architectures, data ratios, and experimental directions will allow the speed of model improvements to break away from the linear pace of human researchers, spinning up the "AI improving AI" evolutionary flywheel.
This is also the outcome Anthropic hopes for the most.
When this flywheel truly starts turning, "AI self-improvement pre-training" will no longer just be a research direction but an accelerated path towards AGI and even ASI.
Currently, all dimensions of competition concerning the arms race of compute power, data barriers, and talent acquisition may be rewritten by this one variable.
Within three years, OpenAI has lost three core figures to the same competitor.
The impact of this fact may be larger than any financing number.
Compute power can be bought with money; data can be accumulated over time, but those who can get the AI evolutionary flywheel moving can be counted on one hand worldwide.
Karpathy chose to put down his freedom identity and return to the front lines at this moment. He believes the window is right in front of him.
免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。
