Explore a 41-minute tutorial demonstrating how to exploit Large Language Model (LLM) vulnerabilities through linguistic hacks and prompt manipulation. Progress through nine levels of doublespeak.chat challenges while learning about prompt leakage, prompt injection, and techniques for circumventing LLM contextual sandboxing. Master the art of discovering hidden secrets by either playing within or manipulating conversation parameters to guide LLM behavior beyond intended boundaries. Access comprehensive resources including GitHub repositories, documentation for various security tools, and detailed write-ups suitable for beginners in cybersecurity and penetration testing. Follow along with clearly marked chapters covering fundamental concepts like jail-breaking LLM sandboxes, reverse prompt engineering techniques, and practical challenge solutions.
Jailbreaking ChatGPT-Style Sandboxes - A Guide to Linguistic Hacks and Prompt Engineering
Overview
Syllabus
Start:
Jail-breaking LLM Sandboxes:
Prompt Leak/Injection:
Reverse Prompt Engineering Techniques:
Forces Unseen: Doublespeak:
Level 1:
Level 2:
Level 3:
Level 4:
Level 5:
Level 6:
Level 7:
Level 8:
Level 9:
End:
Taught by
CryptoCat
