English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

The success of reinforcement learning from human feedback (RLHF) in languagemodel alignment is strongly dependent on the quality of the underlying rewardmodel. In this paper, we present a novel approach to improve reward modelquality by generating synthetic preference data, thereby augmenting thetraining dataset with on-policy, high-quality preference pairs. Motivated bythe promising results of Best-of-N sampling strategies in language modeltraining, we extend their application to reward model training. This results ina self-training strategy to generate preference pairs by selecting the best andworst candidates in a pool of responses to a given query. Empirically, we findthat this approach improves the performance of any reward model, with an effectcomparable to the addition of a similar quantity of human preference data. Thiswork opens up new avenues of research for improving RLHF for language modelalignment, by offering synthetic preference generation as a solution to rewardmodeling challenges.

West-of-N: Synthetic Preference Generation for Improved Reward Modeling