Prompting Challenge by Rahul Parwal

Hello, testers!

This is a fun learning challenge for you to apply the learnings on prompting and prompt engineering. For this challenge, you will have to trick Gandalf (An AI Bot) into revealing secret information – specifically, the password for each level. 

However, the trick is that Gandalf will also level up like you and become more resistant to giving away the password as you progress! 

Here’s how you can approach this challenge:

1. Visit the Gandalf Website: Go to the Gandalf Challenge website and start your adventure.
2. Understand Prompt Injection: The challenge models a real-world problem faced by large language models (LLMs) called prompt injection. Similar to SQL injection attacks, prompt injection occurs when user input (the “data”) mixes with the model’s instructions (the “code”), allowing attackers to manipulate the system. Unlike SQL, LLMs work with natural language, making it challenging to escape input properly.
3. Learn about the LLM Defence Team: Gandalf has set up defenses to prevent Gandalf from revealing the secret password. Pay attention to its strategies and how it protects the password.
4. Think Like a Tester: Now it’s your turn! Try different approaches to trick Gandalf. Experiment with prompts, context, and creative input. Observe how Gandalf responds and adapts your strategy. 

5. Document Your Approach: For each level, document the prompts you used, any insights you gained, and the password you discovered. Screenshots can also help illustrate your process.
6. Submit Your Answer: Create a Google Doc with your findings and grant read access to parwalrahul94@gmail.com. Share the document in the response to this challenge.
7. Post a comment: After you submit your answer, let us know in the comment section down below.

Good luck, and may your prompts be ever clever!