ZerotoMastery – The Dark Side of AI: Jailbreaking, Injections, Hallucinations & more

What you’ll learn

• What jailbreaking models involves and how to do it yourself
• Understanding vulnerabilities inherent to models, including prompt and data leakage
• The risks of exposing LLMs to proprietary or sensitive data
• Exploring the toxicity and bias inherently built into different models
• Real-world tests using ChatGPT, DeepSeek and other models
• Experiment with steering an LLM’s neurons to prevent hallucinations

Don’t get us wrong. AI is awesome. In fact, our instructors teach you plenty about just how awesome it is. But understanding AI’s potential isn’t enough – you need to grasp its pitfalls too. It also has vulnerabilities that can be exploited by people and lead to unintended consequences.

This section takes you through a fascinating deep dive into those risks: jailbreaking, prompt injections, hallucinations, prompt and data leakage, and so much more.

Through real-world demos, cutting-edge models like ChatGPT DeepSeek, and research-backed insights, you’ll see how these issues were discovered, how they show up in the wild, and why even seasoned AI Engineers and Prompt Engineers might be caught off guard.