#compliance-ai News & Analysis

Robust Explanations for User Trust in Enterprise NLP Systems

Researchers propose a black-box robustness evaluation framework for NLP explanations, revealing that decoder-based LLMs produce 73% more stable explanations than encoder models like BERT. The study establishes practical cost-robustness tradeoffs that help organizations select models for compliance-sensitive applications before deployment.