Tag: AI alignment
Browse our exclusive articles!
Mitigating AI Misalignment Contagion with Implicit Steering
Learn how steering with implicit traits helps prevent misalignment contagion in multi-agent AI systems, ensuring safer and aligned interactions.
Safety in Agentic AI Depends on Interaction Topology
Discover why safety and fairness in agentic AI rely on interaction topology, not model scale or alignment, for robust multi-agent decision-making.
Disentangled Preference Optimization: Preserve Winners, Suppress Losers
Discover a novel method to optimize AI preferences by preserving winners and suppressing losers, enhancing large language model alignment and performance.
Localizing and Controlling Policy Circuits in Language Models
Explore how policy routing circuits in language models are localized, scaled, and controlled to enhance safety and performance across model sizes.
Why Refusal-Based AI Alignment Evaluation Fails
Explore why refusal-based AI alignment evaluation is flawed and how routing mechanisms impact AI behavior and censorship strategies.