Summary of Backdoorllm: a Comprehensive Benchmark For Backdoor Attacks on Large Language Models, by Yige Li et al.
BackdoorLLM: A Comprehensive Benchmark for Backdoor Attacks on Large Language Models
by Yige Li, Hanxun Huang, Yunhan Zhao, Xingjun Ma, Jun Sun
First submitted to arxiv on: 23 Aug 2024
Categories
- Main: Artificial Intelligence (cs.AI)
- Secondary: None
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 introduces BackdoorLLM, a comprehensive benchmark for studying backdoor attacks on Generative Large Language Models (LLMs). The benchmark features a repository of backdoor benchmarks with a standardized training pipeline, diverse attack strategies including data poisoning and weight poisoning, extensive evaluations across 7 scenarios and 6 model architectures, and key insights into the effectiveness and limitations of backdoors in LLMs. The authors hope that BackdoorLLM will raise awareness of backdoor threats and contribute to advancing AI safety. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper is about making sure language models are safe from bad attacks. Right now, these models can be tricked into saying things they shouldn’t by using special triggers in the prompt. This has happened before on images and text classification tasks, but not much research has been done on language generation. The authors created a tool called BackdoorLLM that helps test these attacks and figure out how to prevent them. They did lots of experiments with different attacks and models to see what works best. |
Keywords
» Artificial intelligence » Prompt » Text classification
