Dataset · rubend18
Collection of jailbreak prompts for ChatGPT
rubend18/ChatGPT-Jailbreak-Prompts
A small set of ChatGPT jailbreak prompts for studying constraint bypassing.
What's inside
The dataset contains a collection of prompts related to ChatGPT jailbreaking. Presented in tabular and text formats; language is English (metadata also specifies aa). Size is under 1k examples. The card describes it as a resource for understanding and generating text in the context of jailbreaking.
What you can do with it
- Testing LLM robustness to jailbreaks
- Training an unsafe prompt classifier
- Red teaming chat models
- Tasks
- Question answering, Text generation, Fill mask, Zero shot classification, Table question answering
- Languages
- English, Afar
- Size
- up to 1K rows
- Downloads this month
- 2K
- Likes
- 309
Models trained on it
- protectai/deberta-v3-base-prompt-injectionText classification 30.3K
- testsavantai/prompt-injection-defender-large-v0Text classification 5K
- katanemo/Arch-GuardText classification 396
- testsavantai/prompt-injection-defender-small-v0-onnxText classification 208
- testsavantai/prompt-injection-defender-base-v0Text classification 183
- katanemo/Arch-Guard-cpuText classification 105
- testsavantai/prompt-injection-defender-base-v0-onnxText classification 86
- testsavantai/prompt-injection-defender-base-v1-onnxText classification 69
How to load it
from datasets import load_dataset
ds = load_dataset("rubend18/ChatGPT-Jailbreak-Prompts")