Toward Culturally Aligned LLMs through Ontology-Guided Multi-Agent Reasoning 事件

Toward Culturally Aligned LLMs through Ontology-Guided Multi-Agent Reasoning arXiv:2601.21700v3 Announce Type: replace Abstract: Large Language Models (LLMs) increasingly support culturally sensitive decision making, yet often exhibit misalignment due to skewed pretraining data and the absence of structured value representations. Existing methods can steer outputs, but often lack demographic grounding and treat values as independent, unstructured signals, reducing consistency and interpretabili