Andrej Karpathy

Readers usually search Andrej Karpathy for one of three jobs: learning neural networks from first principles, understanding how AI changes software development, or tracking where a respected AI researcher is working. A useful page should not treat him only as a resume entry. It should explain why his tutorials, code repos, terminology, and career moves are used as evidence in LLM, agent, and AI education discussions.

• Learning intent: micrograd, nanoGPT, Zero to Hero, CS231n, LLM101n, and LLM explainers.
• Concept intent: Software 2.0, Software 3.0, vibe coding, agents, and human-in-the-loop engineering.
• Industry intent: OpenAI, Tesla Autopilot, Eureka Labs, and Anthropic pre-training research.

Karpathy is unusually visible because his educational work is executable. micrograd teaches backpropagation through a tiny autograd engine. nanoGPT shows a compact GPT training codebase. Neural Networks: Zero to Hero turns neural-network fundamentals into lecture notebooks and videos. LLM101n and Eureka Labs extend that pattern toward AI-native education, where a human-designed course can be supported by an AI teaching assistant.

Karpathy's Software 2.0 essay made the case that neural networks are not just another model class, but a different way to create programs through datasets, architectures, optimization, and evaluation. His later Software 3.0 talks and community discussion connect that framing to LLMs, prompts, agents, and AI-assisted programming. For GetLLMs, the practical takeaway is that his name often signals a broader shift in who writes the program: a human writing code, a training process producing model weights, or a human directing an LLM/agent through language and context.

Career status is time-sensitive. As of the May 2026 sources used here, Karpathy had joined Anthropic and was associated with pre-training research. Use his own site and primary accounts for long-lived biography and education links, but verify current employment, role, project status, and any claims about Eureka Labs or Anthropic with fresh sources before making time-sensitive recommendations.