Summary of Foot in the Door: Understanding Large Language Model Jailbreaking Via Cognitive Psychology, by Zhenhua Wang et al.
Foot In The Door: Understanding Large Language Model Jailbreaking via Cognitive Psychology
by Zhenhua Wang, Wei Xie, Baosheng Wang, Enze Wang, Zhiwen Gui, Shuoyoucheng Ma, Kai Chen
First submitted to arxiv on: 24 Feb 2024
Categories
- Main: Computation and Language (cs.CL)
- Secondary: Artificial Intelligence (cs.AI)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary A recent surge in Large Language Models (LLMs) has made it possible for people to acquire new knowledge with ease. However, researchers have found that these models are vulnerable to “jailbreaking” attacks, which allow attackers to access restricted information by manipulating the model’s decision-making process. This study sheds light on the intrinsic decision-making mechanism within LLMs when confronted with such jailbreak prompts. The authors propose a psychological explanation based on cognitive consistency theory, suggesting that the key to successful jailbreaking is guiding the LLM to achieve cognitive coordination in an erroneous direction. They also develop an automatic black-box jailbreaking method using the Foot-in-the-Door (FITD) technique, which progressively induces the model to answer harmful questions via multi-step incremental prompts. The study’s prototype system demonstrates a high success rate of 83.9% across eight advanced LLMs. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Large Language Models are amazing tools that can help people learn new things. But, some clever hackers have figured out how to trick these models into giving them secret information. This is called “jailbreaking.” The researchers in this study wanted to understand why the models are so easy to trick and how the hackers do it. They found that the key is to make the model think something that’s not true, like a puzzle or a game. They even developed a way to automatically jailbreak these models using a clever technique called Foot-in-the-Door. This method works by giving the model small steps to follow, each one leading to the next, until it finally gives in and answers the tricky question. |
