Tag: AI alignment
Browse our exclusive articles!
Process Reward Models for Large Language Models Survey
Explore Process Reward Models in LLMs, from outcome signals to process supervision, enhancing reasoning and alignment in AI systems.
Detecting Alignment Faking in LLMs via Tool Selection
Explore how Tatemae detects alignment faking in large language models by analyzing tool selection to improve AI safety and reliability.
Uncertainty-Aware Reward Discounting to Prevent Reward Hacking
Discover how uncertainty-aware reward discounting reduces reward hacking in reinforcement learning, improving stability and real-world AI alignment.
Meta-Aligner: Optimizing Multi-Objective LLM Alignment
Discover Meta-Aligner, a novel framework for dynamic, bidirectional optimization to align large language models with diverse human values effectively.
AVES-DPO: Reducing Hallucinations in LVLMs with Self-Correction
Discover AVES-DPO, a new self-corrected preference learning method that effectively reduces hallucinations in Large Vision-Language Models (LVLMs).