Overview
Learn how to generate and test custom prompt injection datasets using Spikee in this hands-on tutorial video. Follow along with the installation process, workspace initialization, and exploration of dataset seeds before diving into dataset generation with controlled payload positions and delimiters. Master different spotlighting techniques, utilize plugins like 1337, and create customized datasets tailored to specific use cases. Practice testing target LLMs like GPT4-mini, analyze testing results, and discover how to incorporate system messages into datasets. Through detailed demonstrations across multiple timestamps, gain practical experience with Spikee's comprehensive toolkit for prompt injection testing, including custom instructions, various spotlighting methods, delimiter options, and support for multiple jailbreaks and languages.
Syllabus
00:06 - Installation
01:05 - Initializing the workspace
01:46 - Dataset Seeds
06:40 - Generating a Dataset
09:10 - Controlling Payload Position and Delimiters
13:40 - Choosing Different Spotlighting
15:51 - Plugins 1337
17:50 - Creating a Custom Dataset
19:55 - Testing a Target LLM
26:38 - Analyzing the Results
29:21 - Adding System Messages to the Dataset
Taught by
Donato Capitella