TLDR AI · ENGINEERING & RESEARCH
ROBUSTNESS OF LLM-BASED SAFETY JUDGES
Summary from the TLDR AI newsletter
This study highlights vulnerabilities in LLM-based safety judges, showing that prompt sensitivity and adversarial attacks can significantly impact evaluation reliability.