XHS-SCoRE

LLMs struggle with social alignment, generating biased responses and missing social cues

A new paper reveals that current large language models often fail to align with socially desirable preferences, frequently preferring undesirable responses in domains like bias, safety, and ethics. Researchers developed…