Path: Working well on data projects
Stay consistent with yourself and others
See how the same item drifts across a long batch, and re-anchor yourself before the drift reaches the data.
- Starting level
- Comfortable with the basics
- Time
- 20 minutes
- Public lesson guide
- EN + हिंदी
You will learn to
Notice your own drift across a long batch and re-anchor before it reaches the data.
Why this matters
The same input has to get the same label on the first day and the last. Drift is silent, so it is caught by anchors, not by feeling careful.
Three things to remember
- The same item gets the same label on Monday and on Friday.
- Re-read fixed anchor examples before every sitting, not only on day one.
- When you change a past decision, record what changed and why.
See one example
A long batch of short answers has to be marked complete or incomplete against a fixed list of required facts.
Try the judgement yourself
Both answers respond to the task above. One of them does the job better. Which, and why?
Make something yourself
Write three anchor examples for a task you do repeatedly: one clear pass, one clear fail, and one borderline case with the reading you decided to use.
Read further
The source is optional. The complete teaching and practice are available inside AI2Bharat.
Read the primary or official source: Bai et al., Training a Helpful and Harmless Assistant (arXiv)