Karpathy shares LLM practical tutorial: Treat the model as a 'programmable work interface' Former Tesla AI head and OpenAI co-founder Andrej Karpathy shared a post summarizing his core methodology for using LLMs. He emphasizes that context is king, advocating for treating LLMs as a 'programmable work interface'—providing enough background, breaking tasks into smaller parts, and cross-verifying, rather than using it as a one-off Q&A machine.