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/

Zendy Plus

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

Zendy Tools

Zendy Open

Large language models (LLMs) are trained on vast, uncurated datasets thatcontain various forms of biases and language reinforcing harmful stereotypesthat may be subsequently inherited by the models themselves. Therefore, it isessential to examine and address biases in language models, integratingfairness into their development to ensure that these models do not perpetuatesocial biases. In this work, we demonstrate the importance of reasoning inzero-shot stereotype identification across several open-source LLMs. Accurateidentification of stereotypical language is a complex task requiring a nuancedunderstanding of social structures, biases, and existing unfair generalizationsabout particular groups. While improved accuracy is observed through modelscaling, the use of reasoning, especially multi-step reasoning, is crucial toconsistent performance. Additionally, through a qualitative analysis of selectreasoning traces, we highlight how reasoning improves not just accuracy, butalso the interpretability of model decisions. This work firmly establishesreasoning as a critical component in automatic stereotype detection and is afirst step towards stronger stereotype mitigation pipelines for LLMs.

On The Role of Reasoning in the Identification of Subtle Stereotypes in Natural Language