A collection of resources for budding AI hackers. Note this is an evolving resource, so feel free to contribute via a pull request
This repository is meant to be a collection of resources and tools for anyone wanting to dive into programming their own AI or hacking existing AI systems.
These are publicly available resources that anyone can use or access. These are all intended to be intro and novice friendly.
- GPT Prompt Attack: https://gpa.43z.one/
- This is a great example game of how to carry out prompt attacks
- GPT Crafting Game: https://gpt.43z.one/
- A great challenge for learning how to craft specific prompts for LLMs
- Merlin's Defense: https://mcaledonensis.blog/merlins-defense/
- The original popular prompt attack challenge
- LangChain: https://github.com/hwchase17/langchain
- This is a great library for anyone wanting to make their code compatible across multiple LLMs
- Prompt Injections Explained: https://simonwillison.net/2023/May/2/prompt-injection-explained/
- An explainer on why prompt injection attacks work, and why we're kinda stuck with them forever
- Attacking LLM- Prompt Injection: https://www.youtube.com/watch?v=Sv5OLj2nVAQ
- A video discussing what prompt injections are
- Accidental LLM Backdoor - Prompt Tricks: https://www.youtube.com/watch?v=h74oXb4Kk8k
- A video walking through crafting prompt injections
- Jailbreak example: https://www.youtube.com/watch?v=S7jviw0BgKE
- A video showing all the process and steps in crafting you're own prompt injections
- Compromising LLMs using Indirect Prompt Injections: https://github.com/greshake/llm-security
- A great writeup of how an attacker can use indirect prompt injections to attack systems.