Summary of Towards Understanding the Fragility Of Multilingual Llms Against Fine-tuning Attacks, by Samuele Poppi et al.
Towards Understanding the Fragility of Multilingual LLMs against Fine-Tuning Attacks
by Samuele Poppi, Zheng-Xin Yong, Yifei He, Bobbie Chern, Han Zhao, Aobo Yang, Jianfeng Chi
First submitted to arxiv on: 23 Oct 2024
Categories
- Main: Computation and Language (cs.CL)
- Secondary: Artificial Intelligence (cs.AI); Cryptography and Security (cs.CR); Machine Learning (cs.LG)
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 This paper investigates the safety of Large Language Models (LLMs) in multilingual settings. Researchers have shown that LLMs can be compromised by fine-tuning with a few adversarially chosen examples, known as fine-tuning attacks. The study finds that these attacks generalize across languages, allowing malicious actors to exploit vulnerabilities in multiple languages simultaneously. To address this issue, the authors propose a new method called Safety Information Localization (SIL) to identify safety-related information within model parameters. SIL enables the detection of attacks that can break safety alignment across all languages by modifying only 20% of weight parameters. The paper also provides evidence for the alternative pathways hypothesis, which suggests that freezing safety-related parameters does not prevent fine-tuning attacks. Overall, this research highlights the importance of developing robust methods to safeguard LLMs against malicious use. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This study looks at how safe are language models that can understand many languages? Researchers found that these language models can be easily tricked into doing bad things if someone uses a few special examples. This is called an attack, and it works the same way in different languages. The authors suggest a new way to find where safety information is stored in the model’s code and how to fix it. They show that even small changes to the model can make it vulnerable again. This research shows why we need to be careful when using language models and develop ways to keep them safe. |
Keywords
» Artificial intelligence » Alignment » Fine tuning
