Andrej Karpathy

Andrej Karpathy is a Slovak-Canadian computer scientist and AI researcher, born October 23, 1986 in Bratislava, Czechoslovakia (now Slovakia). He is best known as a founding member of OpenAI, the former director of AI and Autopilot Vision at Tesla AI, the primary instructor of Stanford's CS231n deep-learning course, and the author of the "Software 2.0" essay and the Neural Networks: Zero to Hero YouTube series. As of May 2026, he is the founder and chief executive officer of Eureka Labs, an AI-native education company he announced on July 16, 2024.

At a glance

• Education: Computer science and physics at the University of Toronto (BSc, 2009); University of British Columbia (MSc, 2011); Stanford University (PhD in computer science, 2015, dissertation "Connecting Images and Natural Language" under Fei-Fei Li).
• Current role: Founder and Chief Executive Officer of Eureka Labs, since July 2024.
• Key contributions: founding member and research scientist at OpenAI (2015 to 2017); director of AI and Autopilot Vision at Tesla (2017 to 2022); primary instructor of CS231n: Convolutional Neural Networks for Visual Recognition at Stanford (2015 onward); author of "Software 2.0" (2017); creator of the Neural Networks: Zero to Hero YouTube series.
• X / Twitter: @karpathy
• LinkedIn: Andrej Karpathy
• GitHub: karpathy
• YouTube: @AndrejKarpathy
• Personal site: karpathy.ai

Origins

Karpathy was born in Bratislava in October 1986, when the city was part of Czechoslovakia. His family emigrated to Toronto when he was 15. He completed a bachelor's degree in computer science and physics at the University of Toronto in 2009, then a master's degree at the University of British Columbia in 2011 working on reinforcement learning and motor-control problems.

He moved to Stanford for a PhD in computer science, joining Fei-Fei Li's research group at the Stanford Vision Lab. The doctorate, completed in 2015, was titled "Connecting Images and Natural Language" and sat at the intersection of computer vision and natural-language processing. During the program he held two summer internships at Google working on large-scale feature learning over YouTube video, and a 2015 internship at DeepMind on the deep reinforcement learning team.

Career

Karpathy's first faculty-style role was at Stanford itself: in winter 2015 he and Fei-Fei Li launched CS231n: Convolutional Neural Networks for Visual Recognition, one of the first university courses to teach deep learning end to end. Karpathy was the primary instructor through several iterations. The class grew from roughly 150 enrolled in 2015 to more than 700 in 2017, and the publicly posted lecture videos and course notes became a standard reference for graduate-level deep-learning instruction.

In 2015 he joined OpenAI as a founding member and research scientist, working on deep learning for computer vision, generative modeling, and reinforcement learning. He left in June 2017 to join Tesla as Director of AI and Autopilot Vision, reporting to chief executive Elon Musk. At Tesla he led the perception team for the Full Self-Driving program, presented the architecture of the unified "HydraNet" multi-task vision stack at Tesla AI Day in August 2021, and oversaw the transition through Hardware 2, Hardware 3, and Hardware 4. He announced his departure from Tesla in July 2022.

After a sabbatical Karpathy released the Neural Networks: Zero to Hero YouTube series, a multi-lecture program walking through the from-scratch implementation of micrograd, makemore, and nanoGPT. He returned to OpenAI as a senior research scientist in February 2023, where he reported building a team focused on midtraining and synthetic data generation; his "State of GPT" talk at Microsoft Build 2023 dates from this period. He departed OpenAI again on February 13, 2024.

On July 16, 2024 he announced Eureka Labs, an AI-native education company headquartered in San Francisco. The first product, the LLM101n self-study course, walks students through building a "Storyteller" language model from scratch in approximately 12 lectures. Karpathy is sole founder and chief executive officer.

Affiliations

• Stanford University: PhD student (CS, under Fei-Fei Li), unknown to 2015
• Stanford University: Instructor, CS231n, 2015 to present
• OpenAI: Founding member and Research Scientist, 2015 to 2017
• Tesla AI: Director of AI and Autopilot Vision, 2017-06 to 2022-07
• OpenAI: Senior Research Scientist, 2023 to 2024
• Eureka Labs: Founder and CEO, 2024-07-16 to present

Notable contributions

• CS231n: Convolutional Neural Networks for Visual Recognition (Stanford course, 2015 onward). Designed jointly with Fei-Fei Li and co-instructed in later years with Justin Johnson. The companion lecture notes hosted at cs231n.github.io are still cited as the standard introductory deep-learning curriculum, and the Spring 2017 lecture playlist on YouTube is the canonical recorded version of the course.
• char-rnn and "The Unreasonable Effectiveness of Recurrent Neural Networks" (May 2015). A character-level RNN demo and accompanying blog post; one of the early viral demonstrations that neural networks could generate plausible text and code.
• "Software 2.0" (Medium, November 2017). The argument that neural networks would replace hand-coded logic for tasks where they perform better. The essay became a widely cited reference for the framing of neural networks as a new programming paradigm; Karpathy's companion talk, "Building the Software 2.0 Stack" at the Spark+AI Summit 2018, walks through the framework on stage.
• Tesla Autopilot Vision stack (2017 to 2022). Led the camera-only perception architecture across Hardware 2, Hardware 3, and Hardware 4, including the unified HydraNet multi-task model and the bird's-eye-view occupancy approach previewed at the August 2021 and September 2022 Tesla AI Day events.
• Neural Networks: Zero to Hero (YouTube, 2022 to 2023). A from-scratch lecture series implementing micrograd, makemore, and nanoGPT. The series anchors the @AndrejKarpathy YouTube channel, which surpassed one million subscribers per industry coverage in early 2026.
• LLM101n (Eureka Labs, 2024 onward). The inaugural Eureka Labs course; an undergraduate-level curriculum that walks students through training a "Storyteller" language model end to end in Python, C, and CUDA. The repository was made public alongside the company announcement and remains in active development.
• Subsequent open-source releases. nanochat (October 2025), framed as the from-scratch "ChatGPT that $100 can buy," extended the from-scratch lineage to a small but full-stack training and inference setup.

Investments and boards

Karpathy disclosed a small number of named angel positions following his 2024 departure from OpenAI. The entries below are limited to AI, semiconductors, datacenters, software, and energy.

• Eureka Labs (AI / Software): Founder and CEO, 2024. Approximately $20 million seed-stage capital reported in early-2025 industry coverage; investors included Conviction (Sarah Guo), Sam Altman, and adjacent senior AI angel and venture investors.
• Magic (AI / Software): Investor, 2024. Personal angel position announced by Magic co-founder Eric Steinberger in early 2024 following Karpathy's OpenAI departure. Magic builds long-context AI software-engineering systems.

Other angel positions have appeared in third-party investor databases but have not been publicly confirmed by Karpathy or the named companies as of May 2026.

Network

Karpathy's longest-standing professional relationship is with Fei-Fei Li, his Stanford PhD advisor, with whom he co-designed CS231n and continued to share educational content channels. Justin Johnson was a Stanford lab-mate, CS231n co-instructor, and frequent research collaborator; Johnson later co-founded World Labs with Fei-Fei Li in 2024, which keeps the Stanford Vision Lab cohort tightly connected commercially as well as academically.

At OpenAI, Karpathy worked alongside Sam Altman, Greg Brockman, and Ilya Sutskever across both tenures; Altman is also a disclosed Eureka Labs investor. At Tesla, Karpathy reported directly to Elon Musk for five years; the public relationship has been polite but distant since the 2022 departure, in contrast to Musk's litigation against the post-Karpathy OpenAI.

Position in the field

Karpathy is unusual among public AI figures in that his reputation rests on three distinct kinds of work at once: research output (computer vision, language modeling), operating roles at frontier organizations (OpenAI, Tesla AI), and educational artifacts (CS231n, the YouTube series, the from-scratch GitHub repositories).

The educational footprint is the more publicly visible component of that reputation as of May 2026. The @AndrejKarpathy YouTube channel surpassed one million subscribers per industry coverage in early 2026, and his X account at @karpathy carries roughly 2.3 million followers. Industry coverage routinely identifies him as the principal explainer-voice for transformer-era neural networks; TIME named him to its first "100 Most Influential People in AI" list in 2024. Comparators in the public-AI-explainer space include podcaster Lex Fridman and YouTube researcher Yannic Kilcher, though Karpathy's profile rests on instructional rather than interview content.

Eureka Labs's commercial trajectory is a separate question. The company has stayed in early-stage build mode through 2024 to 2026 with limited public-product output beyond LLM101n previews, and industry coverage has noted the gap between Karpathy's founder-credibility-implied expectations and visible Eureka Labs commercial output.

Outlook

Open questions over the next 6 to 18 months:

• LLM101n release cadence. The course has remained a public-development project since the July 2024 founding announcement; a complete release or a defined cohort schedule is the principal commercial signal for Eureka Labs.
• Eureka Labs platform scope. Whether the company evolves into a multi-course AI-native curriculum, a B2B AI-tutor platform, or a single-flagship-course business.
• Continued personal output. The cadence of YouTube releases following nanochat, and the topic mix between technical from-scratch series and broader public talks (the 2025 YC AI Startup School talk, the Dwarkesh podcast appearance).
• Investor pattern. Whether the disclosed Magic position grows into a regular angel cadence comparable to other senior-engineer-turned-investor profiles.
• Public commentary on AGI and agents. Karpathy's "AGI is still a decade away" framing in late-2025 podcast appearances put him at the more conservative end of public timelines; whether and how that position evolves is itself editorially material.
• Potential return to a frontier lab. No public indication of one as of May 2026, but the 2017 to 2024 pattern of moves between OpenAI, Tesla, and OpenAI again leaves the option live if Eureka Labs's trajectory plateaus.

Sources

• Andrej Karpathy. Wikipedia biographical entry; covers career, education, and recognition.
• Andrej Karpathy. Stanford Computer Science faculty page; PhD-era research and CS231n details.
• karpathy.ai. Personal site with current bio, projects, and talks.
• CS231n: Convolutional Neural Networks for Visual Recognition. The Stanford course.
• Software 2.0. The November 2017 Medium essay.
• Neural Networks: Zero to Hero. The YouTube series landing page.
• LLM101n. The Eureka Labs inaugural course repository.
• After Tesla and OpenAI, Andrej Karpathy's startup aims to apply AI assistants to education. TechCrunch coverage of the Eureka Labs founding.
• Ex-OpenAI and Tesla engineer Andrej Karpathy announces AI-native school Eureka Labs. VentureBeat coverage of the founding.
• Andrej Karpathy: The 100 Most Influential People in AI 2024. TIME profile from the 2024 list.
• Eureka Labs founding tweet (July 2024). Karpathy's public announcement of Eureka Labs.
• Photo: Wikipedia entry on Andrej Karpathy, CC-BY 3.0 Gladwin Analytics (Stanford lecture, December 2019).