Ignore This Title and HackAPrompt: Exposing Systemic Vulnerabilities of LLMs through a Global Scale Prompt Hacking Competition
Paper
•
2311.16119
•
Published
•
2
HackAPrompt 1.0: our first global prompt hacking competition
HackAPrompt 1.0 challenged thousands to expose LLM vulnerabilities with 600K+ prompts, yielding an award-winning paper & taxonomy of 29 techniques.
Note At EMNLP 2023, our paper "Ignore This Title and HackAPrompt" shared what we learned from HackAPrompt 1.0. One of our key results was the creation of the Taxonomical Ontology of Prompt Hacking Techniques, a systematized list of 29 different ways people can manipulate language models.
hackaprompt/hackaprompt-dataset
Viewer • Updated • 602k • 462 • 58Note We've released an anonymized dataset of user submissions in the HackAPrompt dataset. This is a treasure trove for anyone interested in studying AI behavior or building stronger defenses.
71hackaprompt
🚀Test prompts and evaluate their effectiveness
Note HackAPrompt Playground is a Hugging Face space that allows you to test your prompt hacking skills!
2hackaprompt
🚀Evaluate prompts against text models and get completions
Note Updated version of our HackAPrompt Playground.