Andrej Karpathy leaves OpenAI to focus on education projects

Andrej Karpathy has left OpenAI. One of the company’s best-known technical figures and an early member of the organization, the engineer said on February 13 that the decision was personal and not prompted by any specific incident.

Karpathy said he now wants to focus on a personal project, work at the intersection of AI and education, and spend more time with his family. His departure comes just a year after he returned to OpenAI, as the company works to consolidate its technical leadership following the launch of ChatGPT and the turbulent governance crisis it faced last November.

A return that lasted a year

Karpathy was part of OpenAI’s founding team in 2015. During that first stint, he worked on applied research in deep learning, the technique neural networks use to learn patterns from large amounts of data.

In 2017, he joined Tesla to lead its artificial intelligence and computer vision efforts. There, he became a prominent voice in the company’s push to develop camera- and neural-network-based assisted and autonomous driving systems. He left Tesla in 2022.

Karpathy announced his return to OpenAI in February 2023. His comeback carried symbolic weight: one of the researchers who helped build the organization before it became a central player in the race to develop language models was rejoining.

His stay, however, has been brief. Karpathy has said that nothing extraordinary happened and that the decision was personal—a significant detail given the scrutiny surrounding OpenAI since its board temporarily ousted Sam Altman in November. Altman returned to the job a few days later, and the company began reorganizing its board.

A leading educator for developers

His importance is not explained solely by his roles at OpenAI or Tesla. Karpathy has become one of the most influential technical educators in modern AI. His classes and videos explain how neural networks and large language models are trained, starting from their basic components.

Earlier this year, he published a series of videos in which he recreated a small language model inspired by GPT-2. The goal was not to compete with OpenAI’s commercial systems, but to teach the process: preparing text, training a neural network and generating new sentences from the patterns it has learned.

That approach has practical value. Models such as GPT-4 are trained using infrastructure and datasets beyond the reach of most people, but understanding their principles allows students, programmers and companies to distinguish between what a model actually does and the exaggerated expectations that often surround it.

OpenAI loses a public figure, not necessarily a critical capability

The departure of a prominent engineer does not, by itself, amount to a strategic rupture. OpenAI has a much larger workforce than it did in its early years, and its products rely on broad research, infrastructure, safety and product teams—not on any one person.

But Karpathy represents a rare kind of profile: researchers who can work with advanced models while also explaining their foundations clearly to a broad community. Because of that combination of industry experience and a commitment to teaching, his next project is likely to draw interest.

For OpenAI, the immediate challenge remains maintaining the pace of model development and restoring institutional stability. For the AI ecosystem, his departure raises the possibility that one of its most widely heard technical communicators will develop his personal project outside a major company.