Tag: AI alignment
Browse our exclusive articles!
Ensuring AI Goal Integrity with Separation-of-Powers Design
Discover the PEA architecture for AI safety, enforcing goal integrity through intent verification, goal drift detection, and formal system-level controls.
Logic Jailbreak: Bypass LLM Safety with Formal Logic
Discover how Logic Jailbreak uses formal logical expressions to efficiently bypass LLM safety restrictions across multiple languages.
Preventing AI Catastrophes: Risks of Misaligned Objectives
Explore the risks of AI misalignment and how constraining capabilities can prevent catastrophic outcomes in advanced AI systems.
HiPO: Enhancing LLM Reasoning with Hierarchical Optimization
Discover HiPO, a novel hierarchical preference optimization method that improves reasoning and alignment in large language models for complex tasks.
Verbal Tics in Large Language Models: Systematic Study
Explore the rise of verbal tics in top LLMs, their impact on naturalness, and the need for improved alignment to enhance AI-human communication.