February 15, 2024
This way of thinking about programming with LLMs is important. It snapped into focus for me, personally, when we started trying in earnest to build conversational experiences that are steerable, incorporate multiple jobs to be done, and need to flexibly access external systems. https://t.co/xWL4Uis3jE
one helpful analogy is to treat the llm something like a cpu or execution device, which takes data and instructions in its context window, and processes information accordingly - karpathy talks about this model a lot