MedEinst: Benchmarking the Einstellung Effect in Medical LLMs
💡 Dataset Description
We are releasing MedEinst under the CC-BY 4.0 license. It is a novel, large-scale counterfactual benchmark explicitly designed to evaluate the Einstellung Effect in medical Large Language Models (LLMs). The dataset contains synthetically modified patient narratives grounded in the proprietary medical knowledge base from the DDXPlus dataset.
While existing benchmarks primarily assess… See the full description on the dataset page: https://huggingface.co/datasets/zhui711/MedEinst.
