Can I read D P: W A D M K ?: Iverging References Hen Do Nnotators Isagree and Do Odels NOW on EtoBox?

D P: W A D M K ?: Iverging References Hen Do Nnotators Isagree and Do Odels NOW by Pranjal Dey is a document available to read on EtoBox.

What is D P: W A D M K ?: Iverging References Hen Do Nnotators Isagree and Do Odels NOW about?

This document investigates diverging preferences in human-labeled preference datasets, identifying ten categories of disagreement sources across four classes: task underspecification, response style, refusals, and annotation errors. The authors reveal that most disagreements stem from individual preferences rather than errors, challenging existing reward modeling methods that treat dissent as noise. They propose new methods for modeling user preferences as distributions to improve pluralistic alignment in l

Author

Pranjal Dey

Language

EN