Skip to content- Sources: primary, discussion
- Summary: The essay argues for learning to code alongside model use. The thread turns on whether a prompt stands to a model as source stands to a compiler.
- Why it matters: If a prompt change can only be evaluated empirically, prompts cannot be reviewed the way source changes are, and review practice has to rest on measured output instead.
- Comments: The strongest objection in the thread is that the compiler analogy fails, because source-to-binary changes can be reasoned about with formal precision while the effect of adding or removing a word in a prompt can only be found by running the model.