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 Tools annual

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 Tools monthly

Global: Zendy Plus monthly

Presents the access of premium content as premium feature

Premium Content

Global: Zendy Plus annual

Global: Zendy Free Plan

Foundation models, i.e. large neural networks pre-trained on large textcorpora, have revolutionized NLP. They can be instructed directly (e.g.(arXiv:2005.14165)) - this is called hard prompting - and they can be tunedusing very little data (e.g. (arXiv:2104.08691)) - this technique is calledsoft prompting. We seek to leverage their capabilities to detect policyviolations. Our contributions are: We identify a hard prompt that adaptschain-of-thought prompting to policy violation tasks. This prompt producespolicy violation classifications, along with extractive explanations thatjustify the classification. We compose the hard-prompts with soft prompt tuningto produce a classifier that attains high accuracy with very littlesupervision; the same classifier also produces explanations. Though thesupervision only acts on the classifications, we find that the modifiedexplanations remain consistent with the (tuned) model's response. Along theway, we identify several unintuitive aspects of foundation models. Forinstance, adding an example from a specific class can actually reducepredictions of that class, and separately, the effects of tokenization onscoring etc. Based on our technical results, we identify a simple workflow forproduct teams to quickly develop effective policy violation detectors.

Using Foundation Models to Detect Policy Violations with Minimal Supervision