Prompt engineering

A language model's output is entirely determined by its input: the prompt. Prompt engineering is the work of finding phrasings, structures, and context that consistently steer the model toward the right answer. It matters because model behavior is sensitive and non-obvious — a single word change can flip a correct answer to a wrong one, or expose a capability the default phrasing hides. The hard part is that this sensitivity is largely opaque: there is no debugger, no principled theory of which prompts work, and results often don't transfer across model versions.