Karpathy加入Anthropic！從共創OpenAI到跳槽，他會帶給Claude什麼改變

The former OpenAI co-founder Karpathy has joined Anthropic, leading Claude’s pre-training research. As talent losses loom at OpenAI’s top ranks, Anthropic is actively poaching top talent, underscoring the intense talent race in the AI industry.

Andrej Karpathy joins Anthropic, responsible for pre-training tasks

OpenAI co-founder, former Tesla AI director, and father of Vibe Coding Andrej Karpathy has announced that he is joining Anthropic.

Anthropic disclosed that this week he has started leading the team to accelerate Claude’s pre-training research. “Pre-training” refers to large-scale training that gives Claude core capabilities and is also the most expensive and computationally intensive phase in building state-of-the-art AI models.

Image source: X Andrej Karpathy joins Anthropic, responsible for pre-training tasks

This major personnel shift in the AI industry shows that as Anthropic’s valuation may surpass OpenAI’s, the company’s efforts to recruit talent have grown even more intense.

One of xAI’s founding members, Ross Nordeen, announced earlier this month that he was joining Anthropic. On the same day, Anthropic reached an agreement with Elon Musk’s SpaceX to lease computing capacity from xAI.

• Related report: Who’s taking over the AI crown? Rumors say Anthropic may raise funds again at a valuation of $90B, surpassing OpenAI

Elon Musk praised Karpathy, revealing insider details from when he was poached at Tesla

Karpathy left OpenAI in 2017 to become Tesla’s AI director, leading the computer-vision team for automated driver-assistance.

While serving as a director at both Tesla and OpenAI, Musk recruited Karpathy to Tesla—a detail mentioned in a lawsuit in which OpenAI CEO Sam Altman prevailed.

Documents submitted to the court show that Musk described Karpathy as the second most important figure in computer vision worldwide, behind OpenAI’s other founder and computer scientist Ilya Sutskever.

Musk said at the time about the recruitment: “The people at OpenAI would want to kill him, but this is something that has to be done.”

Karpathy returns to the front line of AI R&D after Tesla and entrepreneurship

Karpathy was also one of the employees seconded from OpenAI to Tesla to work there “for free,” but at that time, Tesla’s progress in developing autonomous driving vehicles was not as fast as expected.

He left Tesla in 2022, and to this day, the company has still not sold a vehicle that is able—under fully driverless conditions—to be ready to take over at any moment or to drive safely when braking.

After leaving Tesla, Karpathy briefly returned to OpenAI, and then founded the AI education startup Eureka Labs.

Last year, Karpathy proposed the concept of Vibe Coding, which quickly became one of the most popular terms in the AI community.

Translated literally, Vibe Coding means “vibe coding.” It means users mainly think about innovative ideas, instruct an AI coding assistant using prompts, and then generate the way to write the code for the websites or platform systems they want.

• Related report: AI helps solo ventures rise! “Vibe Coding” overturns tradition, and small teams can earn over a hundred million a year

In addition to proposing Vibe Coding, Karpathy also combined an AI code-writing agent with a small language model, conducting 700 experiments within 2 days, reducing large-model training time by 11%. This automated research approach is known in the industry as the Karpathy Loop.

OpenAI top talent moves to the exits; Anthropic strengthens cybersecurity to meet the challenge

As IPO competition between OpenAI and Anthropic grows increasingly intense, Anthropic also recruited Chris Rohlf, who has 20 years of experience, to join its front-line red team for stress testing. He previously worked at Yahoo and Meta and participated in Georgetown University’s CyberAI project.

At the same time, OpenAI has recently lost several senior executives, including John Schulman, who moved to Anthropic in 2024, Sutskever, and Mira Murati, CEO of Thinking Machines.

Anthropic’s founding was also originally closely tied to OpenAI. The company was founded in 2021 by seven former OpenAI employees, including Dario Amodei and his sister Daniela Amodei, who currently serve as Anthropic’s CEO and president, respectively.

Further reading:
Want to buy Claude stock? Anthropic: Unauthorised equity transfers are invalid, and it has sued secondary-market intermediaries