Who Is Andrej Karpathy?
Born in Bratislava, Slovakia in 1986, Karpathy moved to Toronto with his family at age 15. He studied Computer Science and Physics at the University of Toronto, earned a Master's from the University of British Columbia, and completed his PhD at Stanford in 2015 under Fei-Fei Li at the Stanford Vision Lab. His doctoral work focused on the intersection of computer vision and natural language processing, a combination that would define the decade that followed.
From there, his career ran through four of the most important addresses in AI: OpenAI, Tesla, Eureka Labs, and now Anthropic. At each stop, he did something most researchers at that level skip entirely. He explained his work in public. His blog posts, GitHub projects, and YouTube lectures reached engineers who never set foot in a research lab. A generation of AI practitioners trace their first real encounter with deep learning back to something Karpathy posted online for free.
He has nearly 2 million followers on X and is one of the most-watched voices on AI development, research methodology, and what it will actually take to reach artificial general intelligence.
Key Facts
- Born October 23, 1986 in Bratislava (then Czechoslovakia, now Slovakia)
- B.Sc. Computer Science and Physics, University of Toronto (2009)
- M.Sc. University of British Columbia (2011)
- PhD Stanford, supervised by Fei-Fei Li (2015)
- Co-founding research scientist at OpenAI (2015 to 2017)
- Director of AI at Tesla, led Autopilot computer vision (2017 to 2022)
- Brief return to OpenAI (February 2023 to February 2024)
- Founded Eureka Labs, July 2024
- Coined "vibe coding" in February 2025; Collins Dictionary Word of the Year 2025
- Named to TIME's 100 Most Influential People in AI (2024)
- Joined Anthropic's pre-training team, May 2026
The Career That Traced Modern AI
OpenAI: Ground Zero (2015 to 2017)
Karpathy joined OpenAI as one of its founding research scientists immediately after finishing his PhD. OpenAI at that point was a nonprofit, and the founding team included Sam Altman, Ilya Sutskever, Elon Musk, and a handful of researchers who believed the field needed a dedicated, openly published effort to get AI right. Karpathy focused on deep learning and computer vision.
In the same year, he designed CS231n, Stanford's first dedicated deep learning course. The class grew from 150 enrolled students in 2015 to 750 by 2017, making it one of the largest courses on campus. Its lecture notes and problem sets were made freely available online, where they became the de facto entry point into deep learning for an entire generation of engineers. The materials are still in use today.
Tesla: Five Years Building Autopilot (2017 to 2022)
Elon Musk recruited Karpathy while Musk was still a board member at both OpenAI and Tesla. At Tesla, Karpathy served as Director of AI for five years. His team built and maintained the computer vision systems that power Tesla's Autopilot and Full Self-Driving programs, handling everything from in-house data labeling to neural network training and deployment on Tesla's custom inference chip.
At Tesla AI Day in August 2021, Karpathy presented the technical architecture behind Autopilot in unusually granular detail. For a company that guarded its engineering decisions closely, it stood out. His five years at Tesla represent the most direct real-world test of large-scale AI deployment in the industry: not a research paper, but a system used by millions of drivers on public roads.
Back to OpenAI, Then Out Again (2023 to 2024)
Karpathy returned to OpenAI in February 2023, contributing to work around GPT-4 and the ChatGPT product. He left again in February 2024, describing the departure as a personal decision rather than the result of any internal dispute. He moved quickly into independent work from there.
Eureka Labs: Teaching AI to Teach (2024 to 2026)
In July 2024, Karpathy launched Eureka Labs, an AI education company. He framed the long-term vision as an "AI-native school" where expert teachers design the curriculum and AI tutors deliver personalized instruction at scale, comparing it to Starfleet Academy from Star Trek. The first product was LLM101n, a course on large language models. His "Zero to Hero" YouTube series, which teaches neural networks from scratch in code, continued to build a substantial audience throughout this period.
Eureka Labs is currently on hold following his move to Anthropic. Karpathy has said he plans to return to his educational work "in time."
What Is Vibe Coding?
In February 2025, Karpathy posted on X about a style of programming he called "vibe coding." The premise: describe what you want in plain language, let an AI model generate the code, accept changes without reviewing every line, and let the codebase grow organically. His original phrasing was deliberately casual. He said developers should "fully give in to the vibes."
The term spread fast. Collins Dictionary named "vibe coding" its Word of the Year for 2025. Products across the AI developer tooling market, from Cursor to Lovable to Replit, found themselves described through the lens of a phrase that came from a single post. For crypto developers building on-chain applications and DeFi protocols, the shift was tangible: shipping a functional product no longer required deep programming expertise, just a clear description of what you wanted it to do.
Karpathy did not coin the term as a compliment without caveats. He also spent much of 2025 warning that AI agents still produced a lot of "slop" and that the industry was overstating near-term capability. The term was a description of a real behavior, not a blanket endorsement.
The Karpathy Loop: What Is Autoresearch?
In March 2026, Karpathy released autoresearch, a 630-line open-source project that showed, concretely, what autonomous AI research could look like. The setup: give an AI coding agent a small language model, a fixed evaluation metric, and a five-minute compute budget per experiment. Then let it run without anyone in the loop.
Over two days, the agent ran 700 experiments and discovered 20 optimizations independently. When those same improvements were applied to a larger model, training time dropped by 11%. Karpathy described the project as "part code, part sci-fi, and a pinch of psychosis." Others started calling the method the Karpathy Loop.
The significance goes beyond the specific percentage. The experiment proved that AI agents can now close the full research loop for AI training: designing experiments, editing code, collecting results, and stacking improvements without a human making each decision. That project became the direct basis for his role at Anthropic.
Why Andrej Karpathy Joined Anthropic in 2026
On May 19, 2026, Karpathy announced on X that he had joined Anthropic. The post drew nearly 3 million views within the first hour. He joined Anthropic's pre-training team, reporting to Nick Joseph, with a clear mandate: build a team that uses Claude to accelerate the pre-training research that produces the next version of Claude.
Pre-training is the foundational phase of building a large language model. It is the computationally intensive process by which a model absorbs large quantities of data and develops the general knowledge and reasoning it carries into every application built on top of it. Compressing that phase has historically required either enormous amounts of human researcher time or enormous amounts of compute, usually both. Karpathy's team is working to substitute AI for a meaningful share of that human time.
What makes the move worth paying attention to is where Karpathy was standing publicly before it. He had described much of current agentic AI output as "slop." He called reinforcement learning "terrible" for producing genuine reasoning. He estimated AGI was at least a decade away, a timeline he noted was five to ten times more pessimistic than the consensus at AI events in San Francisco. He is not a researcher prone to optimism for its own sake.
His decision to join Anthropic is not a change of view on the pace of general AI progress. It is a narrower, specific judgment: today's models are already capable enough to materially compress the research cycle for the next generation of models, even while falling well short of general intelligence. That distinction matters for how you read his move.
Why Karpathy Matters to Investors and Developers
Karpathy's influence runs at several levels at once. At the research level, he has contributed directly to some of the most consequential AI systems in production: OpenAI's foundational deep learning work, Tesla's Autopilot vision stack, and now Anthropic's pre-training infrastructure. At the cultural level, he has shaped how the industry talks about what AI can and cannot do, through terms like "vibe coding," frameworks like "Software 3.0," and a consistent public skepticism about timelines that carries more weight precisely because of his track record.
At the educational level, his free content has done more to lower the barrier to entry for deep learning than almost any formal institutional effort. CS231n remains widely taught. The Zero to Hero series continues to attract new audiences. The engineers he helped train are now building across the industry, including in crypto, where AI-driven tooling is changing how traders approach the spot market and how protocols are beginning to automate strategies across futures trading.
His career decisions have consistently preceded broader industry shifts rather than followed them. When he moved to Tesla in 2017, real-world AI deployment was still theoretical for most of the field. When he launched Eureka Labs in 2024, AI education infrastructure was still an afterthought. When he built autoresearch in 2026, recursive AI research loops were still considered a distant concept. Tracking what Karpathy works on next is a reasonable proxy for where the frontier is actually heading.
The Bottom Line
Andrej Karpathy has been present at every major inflection point in the current AI era: the founding of OpenAI, the scaling of production AI through Tesla Autopilot, the education infrastructure that trained a generation of practitioners, and now the recursive bet that AI can compress its own development cycle at the pre-training level. His move to Anthropic in May 2026, and the autoresearch work behind it, is the clearest signal yet that self-improving research loops are moving from experimental to active development at the frontier labs.
Learn more about Backpack
Exchange | Wallet | Twitter | Discord | Reddit
Disclaimer: This content is presented to you on an "as is" basis for general information and educational purposes only, without representation or warranty of any kind. It should not be construed as financial, legal or other professional advice, nor is it intended to recommend the purchase of any specific product or service. You should seek your own advice from appropriate professional advisors. Where the article is contributed by a third party contributor, please note that those views expressed belong to the third party contributor, and do not necessarily reflect those of Backpack. Please read our full disclaimer for further details. Digital asset prices can be volatile. The value of your investment may go down or up and you may not get back the amount invested. You are solely responsible for your investment decisions and Backpack is not liable for any losses you may incur. This material should not be construed as financial, legal or other professional advice.
