Autoplay
Autocomplete
Previous Lesson
Complete and Continue
The Dark Side of AI: Jailbreaking, Injections, Hallucinations & more
Introduction
Welcome to The Dark Side (Intro to Guardrails and Jailbreaking) (17:07)
Exercise: Meet Your Classmates and Instructor
Course Resources
The Dark Side of AI
Jailbreak! (The DAN Prompt) (7:25)
Exercise: Create Your Own Jailbreak Prompt
Many Shot Jailbreaking (18:09)
Prompt Injections - Part 1 (9:36)
Prompt Injections - Part 2 (17:42)
Thinking Like LLMs - Multi-Modal Injection (9:17)
Leaking - Part 1 (Prompt Leaking) (8:35)
Leaking - Part 2 (Data Leaking) (18:07)
Exposure (5:40)
Poisoning (3:18)
Toxicity (4:39)
Hallucinations (13:31)
Thinking Like LLMs - Big vs Small (18:58)
Challenge: Conduct Your Own Mechanistic Interpretability Research on Hallucinations (4:34)
Challenge Instructions
Leaderboard: Mechanistic Interpretability
The Model Card (11:05)
Model Cards Deep Dive (14:43)
Exercise: Explore the Model Card for GPT-o3-mini and Learn Something New!
Where To Go From Here?
Let's Keep Learning Together! (0:56)
Review This Byte!
Many Shot Jailbreaking