concept-arc
ConceptARC_Rule_Annotations
ConceptARC-style rule annotations
This release bundles grid-style tasks with natural-language rules written by people or proposed by models, human ratings of how well those rules match the task, and automatic checks of whether each model’s output grid is correct. The is_correct and err fields capture only that grid check, not whether a human endorsed the rule wording. Each row is one attempt on one test case.
On the Hugging Face Hub, the dataset card YAML exposes two subsets (see… See the full description on the dataset page: https://huggingface.co/datasets/AIHumanAbstraction/ConceptARC_Rule_Annotations.bundled-concept-ablation-logsconceptarc-convertedbundled-concept-ablation-v2-logsARC_OWARIDA_conceptConceptARC_Order_flattened