Tag: LLM evaluation
Browse our exclusive articles!
Contrastive Decoding Reduces Score Bias in LLM Evaluations
Learn how contrastive decoding improves Large Language Models' scoring accuracy by reducing score range bias, boosting reliability in LLM evaluations by 11...
ATBench: Realistic Agent Trajectory Benchmark for AI Safety
Discover ATBench, a diverse benchmark for evaluating AI agent safety through realistic multi-step trajectories and risk diagnosis.
Overcoming Self-Preference Bias in LLM Rubric Evaluations
Explore how self-preference bias impacts rubric-based evaluation of large language models and strategies to ensure fair, accurate AI assessments.
Detecting Hallucinations in Mental Health Chatbots Using Human-LLM Hybrid
Enhance mental health chatbot safety by combining human expertise with LLMs to detect hallucinations and omissions in responses accurately.
DOVE: Evaluating LLM Cultural Value Alignment Open-Endedly
Discover DOVE, a novel framework for distributional open-ended evaluation of LLMs' cultural value alignment using a value codebook and optimal transport.