Summary of Mitigating Adversarial Attacks in Llms Through Defensive Suffix Generation, by Minkyoung Kim et al.
Mitigating Adversarial Attacks in LLMs through Defensive Suffix Generation
by Minkyoung Kim, Yunha Kim, Hyeram Seo, Heejung Choi, Jiye Han, Gaeun Kee, Soyoung Ko, HyoJe Jung, Byeolhee Kim, Young-Hak Kim, Sanghyun Park, Tae Joon Jun
First submitted to arxiv on: 18 Dec 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
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 novel algorithm is proposed to enhance the robustness of Large Language Models (LLMs) against adversarial attacks. The gradient-based defensive suffix generation method appends carefully optimized suffixes to input prompts, mitigating harmful influences while preserving model utility. A total loss function combines defensive and adversarial losses to generate effective defensive suffixes. Experimental evaluations on open-source LLMs like Gemma-7B, mistral-7B, Llama2-7B, and Llama2-13B show an average 11% reduction in attack success rates (ASR) compared to models without defensive suffixes. Additionally, the perplexity score of Gemma-7B decreased from 6.57 to 3.93 with the proposed method. TruthfulQA evaluations demonstrate consistent improvements with Truthfulness scores increasing by up to 10% across tested configurations. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper talks about how to make language models safer against fake inputs that can cause bad outputs. A new way to add a special suffix to input prompts helps protect the model from these attacks while still working well. The approach uses a special combination of two types of losses to generate effective defensive suffixes. Tests on several language models show that this method reduces the chance of an attack being successful by 11%. It also makes some models better at generating truthful responses, which is important for using language models in critical applications. |
Keywords
» Artificial intelligence » Loss function » Perplexity
