ML//DSPy

- Stanford NLP framework: program (not prompt) language models — treat prompts as compiled, optimizable programs.

Stanford NLP framework: program (not prompt) language models — treat prompts as compiled, optimizable programs.

Define signatures (input→output), compose modules, optimize via few-shot examples or metrics.

Core idea: humans are terrible zero-shot prompters, so let the system optimize prompts automatically.

Represents the "automatic prompt engineering" direction — prompts as code, not prose.