dataset_info:
features:
- name: text
dtype: string - name: openai_classification
dtype: bool - name: harmful_categories
dtype: string - name: full_response
dtype: string
splits: - name: train
num_bytes: 32005666
num_examples: 22286
download_size: 7938470
dataset_size: 32005666
configs: - config_name: default
data_files:- split: train
path: data/train-*
license: apache-2.0
language:
- split: train
- en
tags: - guardrail
- compliance
- threats
pretty_name: Harmful Questions
size_categories: - 1K<n<10K
Dataset Card for Harmful Dataset Validation
Dataset Details
Description
This dataset is designed for validating dataset guardrails by identifying harmful content. It includes classifications and categories of potentially harmful responses.
- Curated by: ZySec AI
- Language: English
- License: [More Information Needed]
Uses
Direct Use
For testing and improving dataset filtering mechanisms to mitigate harmful content.
Out-of-Scope Use
Not intended for deploying unfiltered models or generating harmful content.
Dataset Structure
Includes text data, classification labels, and harmful category annotations.
Dataset Creation
Curation Rationale
To assess and refine guardrails for LLM-based content moderation.
Annotations
Includes harmful category labeling and classification.
Bias, Risks, and Limitations
Potential biases in classification and category labeling. Users should evaluate before application.
Recommendations
Use with caution and in compliance with ethical AI guidelines.