duke-trust-lab/human-aligned-similarity-benchmark

Human Aligned Similarity Benchmark You are welcome to go to alignedmachine.com to contribute. Overview This dataset contains human-aligned similarity judgments for embedding text and multimodal AI model evaluation. The benchmark is designed to assess how well AI models align with human cognitive preferences in similarity perception across text and image modalities. Dataset Structure Concept Files This dataset contains human preference… See the full description on the dataset page: https://huggingface.co/datasets/duke-trust-lab/human-aligned-similarity-benchmark.