Andrej Karpathy

Andrej Karpathy is a researcher and educator whose career connects neural-network research with large production systems. He was a founding member of OpenAI, later led artificial intelligence at Tesla, and developed Stanford’s deep-learning course during his doctorate. His public teaching makes the mechanics of these systems accessible well beyond a research lab.

Autoresearch brings that interest in fundamentals into a small, inspectable experiment loop. An agent changes a model-training program, runs a five-minute experiment, checks the result, and keeps or discards the change. The project makes the instructions governing the experiments a central part of the research process.

For a developer with suitable computing resources, it offers a way to explore automated training experiments and study which changes help under a fixed time budget. His video lessons provide a complementary starting point: understand how language models learn before deciding what an automated experiment should optimize.

Radar follows Karpathy because he connects clear explanations with working code. Autoresearch is especially interesting as an example of agents doing measurable work, with experiment logs that readers can inspect. Improvements within its training setup should be understood in that context rather than treated as proof of general intelligence.