Human-annotated data used to evaluate LLM judges of authoritarianism in LLM responses.
See https://arxiv.org/abs/2606.16127 for more information.
dataset_info:
features:
- name: judge_model
dtype: string
- name: prompt
dtype: string
- name: model
dtype: string
- name: task
dtype: string
- name: response
dtype: string
- name: label_aggression
dtype: int64
- name: label_submission
dtype: int64
- name: label_conventionalism
dtype: int64
- name: label_refusal
dtype: int64
- name: label_combined
dtype: int64
- name: tag
dtype: float64
- name: predicted_as_autho
dtype: bool
splits:
- name: train
num_bytes: 3184663
num_examples: 296
download_size: 1576799
dataset_size: 3184663
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
license: apache-2.0
task_categories:
- text-classification
language:
- en
