New VIBE benchmark measures affective framing in LLM outputs

Researchers have introduced VIBE, a new benchmark designed for the affective profiling of large language model outputs. This benchmark focuses on entity-centered VAD (Valence-Arousal-Dominance) attribution, separating scalar favorability from response-level and target-directed VAD. The VIBE benchmark includes a measurement contract that distinguishes generation from external scoring and reports profiles through an Affective Passport, emphasizing the need for documented practices in affective profiling. AI

IMPACT Provides a new method for evaluating the affective framing and potential biases within LLM-generated text.

RANK_REASON The cluster describes a new academic benchmark for evaluating LLM outputs. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• Andrey Chetvergov
• arXiv
• Hugging Face
• Valence-Arousal-Dominance
• VIBE

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →