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

Deploying large language models (LMs) can pose hazards from harmful outputssuch as toxic or false text. Prior work has introduced automated tools thatelicit harmful outputs to identify these risks. While this is a valuable steptoward securing models, these approaches rely on a pre-existing way toefficiently classify undesirable outputs. Using a pre-existing classifier doesnot allow for red-teaming to be tailored to the target model. Furthermore, whenfailures can be easily classified in advance, red-teaming has limited marginalvalue because problems can be avoided by simply filtering training data and/ormodel outputs. Here, we consider red-teaming "from scratch," in which theadversary does not begin with a way to classify failures. Our frameworkconsists of three steps: 1) Exploring the model's range of behaviors in thedesired context; 2) Establishing a definition and measurement for undesiredbehavior (e.g., a classifier trained to reflect human evaluations); and 3)Exploiting the model's flaws using this measure to develop diverse adversarialprompts. We use this approach to red-team GPT-3 to discover classes of inputsthat elicit false statements. In doing so, we construct the CommonClaim datasetof 20,000 statements labeled by humans as common-knowledge-true, commonknowledge-false, or neither. We are making code and data available.

Explore, Establish, Exploit: Red Teaming Language Models from Scratch