AI Guardrails Index: Topic Restriction

We broke AI guardrails down to six categories. We curated datasets and models that demonstrate the state of AI safety using LLMs and other open source models.

Topic Restriction

Competitor Check

Hallucination

Content Moderation

PII Detection

Jailbreaking

Introduction

Restricted Topics guardrails ensure AI interactions remain focused and relevant, similar to human professional interactions. They prevent AI systems from straying into unintended domains, politely redirecting off-topic queries. Essential in various applications like customer service, financial advising, and medical information systems, these guardrails enhance user experience, maintain conversation integrity, and minimize risks associated with off-topic discussions across diverse industries.

Data Breakdown

See the full dataset here: Topic Restriction dataset

Conclusion

The Guardrails AI Ensemble model demonstrates superior performance for broad topic restriction, outperforming Google Natural Language in both general and specific classifications. While the ensemble model has higher latency, GPU acceleration significantly mitigates this drawback. The Guardrails AI Ensemble is particularly valuable for critical applications where performance is paramount, such as content moderation and compliance monitoring. For most real-world scenarios, its improved F1 score and comprehensive topic coverage outweigh the slight increase in processing time. When choosing between models, users should carefully balance F1 scores requirements against latency constraints, considering the specific needs of their application. Overall, the Guardrails AI Ensemble emerges as the preferred solution for robust and accurate topic restriction across a wide range of use cases.

Leaderboard