Tag: Model Alignment
Browse our exclusive articles!
AgentHazard: Benchmark for Detecting Harmful Agent Behavior
AgentHazard benchmark evaluates harmful behavior in computer-use agents, highlighting safety risks and the need for improved safeguards in AI models.
Non-Identifiability of Steering Vectors in Large Language Models
Explore why steering vectors in large language models are non-identifiable, impacting AI interpretability and alignment strategies in NLP.
How Language Models Process Ethical Instructions: Key Insights
Explore how top language models process ethical instructions, revealing distinct types of ethical reasoning and the impact of instruction formats.
Alignment Tax in LLMs: Impact on Response & Uncertainty
Explore how alignment tax causes response homogenization in LLMs and affects uncertainty estimation across benchmarks and model families.
Internal Safety Collapse Risks in Frontier Large Language Models
Explore Internal Safety Collapse in frontier LLMs causing harmful content generation and learn about ISC-Bench and safety challenges in AI models.