When in Doubt, Cascade: Towards Building Efficient and Capable Guardrails

TL;DR

This paper presents a scalable, cost-effective approach to developing guardrail models for large language models by using synthetic data generation and systematic evaluation to improve detection of undesirable outputs.

Contribution

The authors introduce a synthetic data pipeline leveraging taxonomy-driven instructions to enhance guardrail models for LLMs, achieving competitive performance efficiently.

Findings

Generated over 300K contrastive samples for training

Achieved competitive guardrail detection performance

Provided insights into iterative model development

Abstract

Large language models (LLMs) have convincing performance in a variety of downstream tasks. However, these systems are prone to generating undesirable outputs such as harmful and biased text. In order to remedy such generations, the development of guardrail (or detector) models has gained traction. Motivated by findings from developing a detector for social bias, we adopt the notion of a use-mention distinction - which we identified as the primary source of under-performance in the preliminary versions of our social bias detector. Armed with this information, we describe a fully extensible and reproducible synthetic data generation pipeline which leverages taxonomy-driven instructions to create targeted and labeled data. Using this pipeline, we generate over 300K unique contrastive samples and provide extensive experiments to systematically evaluate performance on a suite of open source datasets. We show that our method achieves competitive performance with a fraction of the cost in compute and offers insight into iteratively developing efficient and capable guardrail models. Warning: This paper contains examples of text which are toxic, biased, and potentially harmful.