All learning paths Learn See Try Make The guideline decides, not your taste What this lesson is about Apply a written guideline even when you privately disagree, and raise your disagreement the right way.
Your rule book counts a reply refusing a medical question as helpful. You think it helps nobody, and the batch is due today. What now?
A Mark it your way, and change it back if anyone raises it later. B Skip every item like this until somebody answers your question about it. C Where the guideline decides the case, the guideline wins over your taste.
Check my recall Why this matters Your labels are compared with everyone else's. A quiet exception makes the data disagree with itself, and leaves no record of why. Where the guideline names a criterion — one quality you judge by — that criterion decides, not your taste.
Apply a guideline you disagree with, and move the disagreement to where it can change the guideline.
Your labels are compared with everyone else's. A quiet exception makes the data disagree with itself, and leaves no record of why. Where the guideline names a criterion — one quality you judge by — that criterion decides, not your taste.
Where the guideline decides the case, the guideline wins over your taste. Send your disagreement as a separate note, not as a changed label. Quote the guideline line you applied whenever a case was close. Task The guideline treats a short answer as complete when it names all three required facts. A reviewer finds one such answer too blunt.
Weak approach Checks the answer against the guideline and confirms all three required facts are there. Then marks it incomplete, because it reads rudely, and writes nothing anywhere about why. The label now disagrees with everyone else's on the same rule, and nothing says why.
Stronger approach Marks it complete as the guideline states, and files a separate note asking whether tone should become its own criterion.
Why the stronger approach works The label stays comparable with every other worker's label, and the real issue reaches the people who can fix it.
Try a changed situation You are now a volunteer filling forms at a camp, following the office's written rules. Which rule still applies?
Make something yourself Describe a rule you have been asked to follow that you disagreed with. Write how you applied it anyway, and the separate note you would send about the gap.
Lesson 1 of 7 on Working well on data projects next Ask a good question about an edge case Part 1 ends in a work sample you can send
Read the primary or official source: OpenAI InstructGPT labeler guidelines