Why Did Karpathy Suddenly Join Anthropic, Only to Be Dario's “-2”?

Original Editor: Marco

Source: 新智元

At 11 PM on May 19, Andrej Karpathy officially announced his join of Anthropic.

The weight of this name needs no further explanation.

OpenAI co-founder, former AI Director at Tesla, father of "Vibe Coding," and one of the most influential AI educators globally.

His standing in the AI field is roughly equivalent to LeBron James in basketball – his every move makes headlines.

He only posted three sentences on X.

https://x.com/karpathy/status/2056753169888334312

The first sentence said the next few years at the LLM frontier would be "particularly formative." The third said he still loves education. The middle sentence was the most crucial, five words: "return to research."

This marks the third core figure to defect from the OpenAI camp to Anthropic in two years.

It is also someone about to turn 40, who has achieved fame and fortune and financial freedom, actively choosing to be someone else's subordinate.

Why leave? Why Anthropic? And why did Anthropic have to hire him?

Each question has layers worth unpacking.

What He Will Do

Karpathy started work this week, joining the Anthropic pre-training team.

This team is led by Nick Joseph and is responsible for all large-scale training runs for Claude.

An Anthropic spokesperson confirmed to TechCrunch that Karpathy will form a new sub-team focused on using Claude itself to accelerate pre-training research.

Nick Joseph also provided context in a post on X, "He will build a team focused on using Claude to accelerate pre-training research itself."

https://x.com/nickevanjoseph/status/2056760504949842219

TechCrunch commented, "Karpathy is one of the few researchers who can bridge the gap between LLM theory and large-scale training practice."

Axios characterized the move as "a major talent acquisition victory for Anthropic."

On the same day, cybersecurity expert Chris Rohlf also announced he was joining Anthropic, following former xAI founding member Ross Nordeen, who joined earlier this month. The directional flow of talent is becoming increasingly clear.

https://x.com/chrisrohlf/status/2056744653165092983

Data from Polymarket can serve as a barometer of market sentiment – traders price the probability of Anthropic having the best AI model by the end of June at 65%, compared to 4% for OpenAI.

https://polymarket.com/event/which-company-has-best-ai-model-end-of-june

Karpathy's addition further reinforces this assessment.

Karpathy, the Definer

To understand the significance of this move, one must understand Karpathy's unique value.

His uniqueness isn't just technical ability; there are many top-tier researchers.

His uniqueness lies in his ability to change the entire industry's understanding of something with a single term.

Born in Slovakia in 1986, he immigrated to Toronto, Canada, at 15.

During his undergraduate studies at the University of Toronto, he took a course by Geoffrey Hinton and attended his reading group.

Hinton is the spiritual leader of the deep learning renaissance, a 2018 Turing Award winner, and a 2024 Nobel Prize in Physics winner.

Karpathy was one of the first young minds ignited by this flame.

Later, he 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, it became the first stop for countless engineers worldwide teaching themselves deep learning, and remains the quintessential computer vision course.

In 2015, he became a founding research scientist at OpenAI.

In 2017, he was recruited by Elon Musk to Tesla as Senior Director of AI, pushing autonomous driving towards a pure vision approach.

Musk faced immense 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," arguing 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, pushing his channel past a million subscribers.

During the same period, his open-source projects – micrograd, nanoGPT, nanochat – featured minimal code but precisely hit core concepts, earning the nickname "runable textbooks."

In February 2025, he coined the term "Vibe Coding," which was selected as the Word of the Year by the Collins Dictionary.

https://x.com/karpathy/status/1886192184808149383

In June, during a speech at YC AI Startup School, he proposed the frameworks of "Software 3.0" and the "Decade of Agents," becoming one of the most widely discussed AI speeches of the year.

TIME listed him among the "100 Most Influential People in AI" in 2024.

From Hinton to Fei-Fei Li, to Altman, and then to Musk, he has stood at the forefront at every juncture.

But the most enduring things he leaves behind are not any single product or paper, but those conceptual frameworks.

Software 2.0, Vibe Coding, LLM OS. These terms changed how people think about AI.

Why Settle for Being 'Tier -2'?

Karpathy's career has a clear thread: he has never chased titles.

He has been a student of Hinton and Fei-Fei Li, a colleague of Altman, and a direct report to Musk.

In each of these experiences, his position in the organizational chart was at a high level.

Now he joins Anthropic, reporting directly to Nick Joseph, the head of pre-training.

Nick Joseph reports to Dario Amodei.

Karpathy is now at the third tier of the organizational chart.

Nick Joseph is one of Anthropic's 11 founders, previously working at Vicarious and OpenAI.

At OpenAI, he worked on code models in the safety team. Seeing that fine-tuned GPT-3 could write code, he realized AI could self-improve. He then left with the leader of the safety team to found Anthropic.

His team trained the entire Claude model family, including Mythos.

The reason Karpathy is willing to do research under Nick Joseph is simple: this position is closest to the work he wants to do.

Looking back at every career move he has made, the driving force has been the same: "Where is the biggest experiment right now?"

Going to Tesla in 2017 because autonomous driving was the biggest experimental ground for Software 2.0.

Leaving in 2022 because the architecture was set, and only engineering optimization remained.

Returning to OpenAI in 2023 because the explosion driven by ChatGPT following the GPT-4 release was the most exciting frontier.

Founding Eureka Labs in 2024 to test the hypothesis of AI-native education.

Joining Anthropic in 2026 because the pre-training revolution of "using AI to research AI" is happening here.

His departures have never been due to dissatisfaction but because the current location was no longer the site of the biggest experiment.

Why not return to OpenAI? Talent flow provides the answer.

Jan Leike, former head of alignment at OpenAI, joined Anthropic in May 2024.

John Schulman, OpenAI co-founder, followed suit in August of the same year.

Now it's Karpathy's turn.

Three people in two years, all flowing in one direction, with no comparable reverse cases.

OpenAI's strategic focus has shifted from pure research towards platformization and acquisitions. Chat.com, io Products, Windsurf, TBPN – the intervals between acquisitions are getting shorter, and the amounts are getting larger.

This is a company transforming into an "AI-era consumer giant."

For a researcher wanting to "return to research," Anthropic's path of "competing on research quality" is more attractive.

Why Anthropic Wanted Him So Badly

Anthropic's motivation for this hire can be broken down into several layers.

The most superficial layer is technical need.

No matter how large Anthropic's compute budget is, it cannot match Microsoft-backed OpenAI or Google with its TPUs.

Anthropic cannot win a pure compute arms race.

It must find a way to train better models with less compute.

"Using Claude to accelerate pre-training research" is exactly this path, and Karpathy possesses the rare combination of deep pre-training theory, large-scale engineering experience, and intuition for AI-assisted research.

Beneath that is the talent signal.

With three OpenAI core figures flowing unidirectionally to Anthropic in two years, the narrative of "frontline researchers voting with their feet" is now established.

Every addition at Karpathy's level lowers the psychological barrier for the next top-tier talent to join. Talent attracts talent; the flywheel spins itself.

Then there's the brand polishing before an IPO.

Anthropic is negotiating a $30 billion funding round at a $900 billion valuation, and IPO preparations are underway.

Karpathy is one of the most publicly recognizable technical figures in AI: a million YouTube subscribers, a Word of the Year coiner, the maintainer of the 220k-star CLAUDE.md repository on GitHub.

His name on the Anthropic employee list gives bankers a ready-made phrase for the prospectus.

But the most intriguing layer might be something Anthropic didn't explicitly state as a hiring motive but is destined to yield the greatest return: Karpathy's ability to define paradigms.

Any technical exploration he does at Anthropic will be discussed publicly by him – on X, in blog posts, on YouTube videos.

When he names what is happening in his uniquely characteristic way, Anthropic naturally becomes the birthplace of that paradigm.

Hiring a top pre-training researcher comes with the bonus of acquiring the industry's most influential technical narrator.

The Tipping Point of the Flywheel

Placing this personnel change in a broader context reveals a technological inflection point.