Andrej Karpathy
todayFormer director of AI at Tesla and OpenAI founding member. Now one of the field's most-watched educators — his nanoGPT lectures introduced millions to LLM internals from first principles.
- Tesla Autopilot vision stack lead.
- nanoGPT and 'Let's build GPT' YouTube series.
- Karpathy's law of LLM debugging: 'It's almost always the data.'
FAQ
What did Karpathy do at Tesla?
He led the Autopilot vision team from 2017 to 2022, building the neural-network stack that processes camera input across Tesla's fleet. Under his tenure the team shifted from hand-engineered perception modules to a unified, end-to-end deep-learning architecture.
What is nanoGPT?
Karpathy's minimal, single-file PyTorch implementation of a GPT-style transformer. Together with his 'Let's build GPT' lecture series it has become the de facto first read for engineers learning LLM internals from first principles.
What is Karpathy's law of LLM debugging?
'It's almost always the data.' His shorthand for the empirical observation that when an LLM behaves badly, the highest-yield place to look is the training or evaluation dataset — labelling errors, distribution mismatch, or contamination — not the model architecture.
What was Karpathy's role at OpenAI?
He was a founding member of OpenAI in 2015 and rejoined as a research scientist in 2023, working on building agents and training data improvements before departing again in 2024 to focus on independent AI-education work.