Tag: LLM safety
Browse our exclusive articles!
Balancing Ethical Reasoning and Safety in Large Language Models
Explore how TRIAL and ERR frameworks improve ethical reasoning and safety alignment in Large Language Models to prevent harmful outputs.
LatentRefusal: Safe Refusal for Unanswerable Text-to-SQL Queries
Discover LatentRefusal, a lightweight mechanism that improves refusal accuracy in text-to-SQL systems by detecting unanswerable queries safely and efficien...
Benchmarking Outcome-Driven Constraint Violations in AI Agents
Discover a new benchmark assessing outcome-driven constraint violations in autonomous AI agents to improve safety and ethical compliance under KPI pressure...
LASA: Enhancing LLM Safety with Language-Agnostic Alignment
Discover LASA, a novel framework improving LLM safety across languages by aligning semantics at the model's semantic bottleneck for robust protection.
AISafetyBenchExplorer: Unifying AI Safety Benchmarks & Metrics
Explore AISafetyBenchExplorer, a comprehensive AI safety benchmark catalogue revealing fragmented metrics and weak governance in AI safety evaluation.