Summary of Round Trip Translation Defence Against Large Language Model Jailbreaking Attacks, by Canaan Yung et al.
Round Trip Translation Defence against Large Language Model Jailbreaking Attacks
by Canaan Yung, Hadi Mohaghegh Dolatabadi, Sarah Erfani, Christopher Leckie
First submitted to arxiv on: 21 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 The proposed Round Trip Translation (RTT) method is a novel algorithm designed to defend large language models against social-engineered attacks. These attacks require a high level of comprehension for the LLMs to counteract, making existing defensive measures ineffective against most of them. The RTT method paraphrases the adversarial prompt and generalizes the idea conveyed, enabling LLMs to detect induced harmful behavior more easily. This approach is versatile, lightweight, and transferable across different LLMs, successfully mitigating over 70% of Prompt Automatic Iterative Refinement (PAIR) attacks, as well as reducing the success rate of MathsAttack by almost 40%. The code is publicly available on GitHub. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Large language models are vulnerable to social-engineered attacks that are difficult for them to understand. These attacks can be very harmful! Currently, there’s no good way to stop most of these attacks. That’s why scientists came up with a new idea called Round Trip Translation (RTT). This method makes it easier for large language models to detect when someone is trying to trick them into doing something bad. The RTT method works well and can even stop over 70% of the most effective kind of attack. It also helps reduce another type of attack by almost 40%. You can find the code for this new idea on a website called GitHub. |
Keywords
» Artificial intelligence » Prompt » Translation
