Skip to content
Expert Dossiers
Expert chat
Artificial intelligence research and education

Andrej Karpathy

An AI researcher and educator known for deep learning, Tesla Autopilot, OpenAI, CS231n, and unusually clear technical teaching.

Andrej Karpathy's public work spans academic computer vision, OpenAI research, Tesla Autopilot, neural-network education, and software-shaped thinking about AI. This dossier starts from his own site, Stanford pages, major talks, writing, and open-source teaching projects.

Chat with this dossier Browse all dossiers

Why they matter

Karpathy is rare because his credibility crosses research, production systems, and teaching. He can explain the mathematical and systems pieces of deep learning while also showing how they become usable software and public education.

Expertise map

Deep learning and neural network trainingComputer vision and language-vision modelsLLMs and AI educationAutonomous driving perception systemsDeveloper education and from-scratch implementation

Core ideas

Neural networks are trained systems, not hand-coded rules

Karpathy's public writing and talks consistently point to learned behavior from large datasets as a software primitive, most famously in his Software 2.0 framing.

Understand the system by rebuilding the small version

His educational style favors from-scratch implementations and simplified end-to-end builds, visible in Zero to Hero, micrograd, and LLM teaching material.

Teaching is a serious research artifact

CS231n and later YouTube lectures show that his influence is not just papers or jobs; it is the public transfer of deep learning intuition to builders.

Timeline

  1. 2005-2009

    Studied computer science and physics at the University of Toronto, according to his site.

  2. 2011-2015

    Completed Stanford PhD work focused on convolutional/recurrent neural networks and language-vision applications.

  3. 2015-2017

    Worked as a research scientist and founding member at OpenAI, according to his site.

  4. 2017-2022

    Served as Director of AI at Tesla, leading computer vision work for Autopilot.

  5. 2023-2024

    Returned to OpenAI and built a team working on midtraining and synthetic data generation, according to his site.

  6. 2024-present public work

    Creates AI education videos, maintains public writing/projects, and announced Eureka Labs as an AI-native school project.

Fair criticism

  • His teaching can make complex systems feel approachable; users still need to do the math, implementation, and empirical work.
  • Some of his older project pages are explicitly outdated, so dossier freshness needs regular review.
  • Public explanations are not substitutes for safety, deployment, or evaluation expertise in production AI systems.

Beginner path

Karpathy official site

Best map of his career timeline, talks, writing, teaching, and projects.

Open source

Intro to Large Language Models

Accessible public entry point to his LLM explanation style.

A Recipe for Training Neural Networks

A practical, durable guide to training discipline and debugging instincts.

Open source

Advanced questions

  • How does Software 2.0 change the boundary between code, data, and product behavior?
  • Which Karpathy teaching projects best predict real engineering competence?
  • How should builders adapt his from-scratch teaching style to frontier-model application work?

Source trail

officialKarpathy official site

Current self-maintained career timeline, talks, teaching, writing, and project links.

primaryCS231n

Stanford course associated with Karpathy's early teaching influence.

primarySoftware 2.0

Foundational essay for his learned-software framing.

primaryA Recipe for Training Neural Networks

Practical training and debugging guide.

primaryEureka Labs

Announcement for his AI-native school project and LLM101n course direction.

primarymicrograd

From-scratch autograd teaching project.

Risk notes

  • Not affiliated with Andrej Karpathy.
  • AI system advice should be source-backed and distinguish conceptual explanation from production readiness.