Detectors for Safe and Reliable LLMs: Implementations, Uses, and Limitations

TL;DR

This paper discusses the development and deployment of compact detectors for identifying risks in large language models, highlighting their uses, challenges, and future directions for improving AI safety and governance.

Contribution

It introduces a library of simple, effective classifiers for detecting various harms in LLMs and explores their applications and limitations.

Findings

Detectors can identify biases, toxicity, and non-faithful outputs.

They serve as safety guardrails and support AI governance.

Challenges include reliability and scope expansion.

Abstract

Large language models (LLMs) are susceptible to a variety of risks, from non-faithful output to biased and toxic generations. Due to several limiting factors surrounding LLMs (training cost, API access, data availability, etc.), it may not always be feasible to impose direct safety constraints on a deployed model. Therefore, an efficient and reliable alternative is required. To this end, we present our ongoing efforts to create and deploy a library of detectors: compact and easy-to-build classification models that provide labels for various harms. In addition to the detectors themselves, we discuss a wide range of uses for these detector models - from acting as guardrails to enabling effective AI governance. We also deep dive into inherent challenges in their development and discuss future work aimed at making the detectors more reliable and broadening their scope.