Summary of Universal Vulnerabilities in Large Language Models: Backdoor Attacks For In-context Learning, by Shuai Zhao et al.
Universal Vulnerabilities in Large Language Models: Backdoor Attacks for In-context Learning
by Shuai Zhao, Meihuizi Jia, Luu Anh Tuan, Fengjun Pan, Jinming Wen
First submitted to arxiv on: 11 Jan 2024
Categories
- Main: Computation and Language (cs.CL)
- Secondary: Artificial Intelligence (cs.AI); Cryptography and Security (cs.CR)
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 paper explores the security vulnerabilities of in-context learning, a popular AI paradigm for natural language processing (NLP) tasks. Despite its efficacy, in-context learning is susceptible to malicious attacks that manipulate large language models without requiring fine-tuning. The researchers design a new backdoor attack method called ICLAttack, which poisons demonstration context or prompts to control model behavior. This method preserves the model’s generality and achieves an average attack success rate of 95.0% across multiple datasets on OPT models. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary In-context learning helps AI learn from few examples by showing them how to solve a problem. However, this approach has a secret weakness: it can be tricked into doing what attackers want without needing to retrain the model. The researchers found that by tampering with the “demonstration” data used during training, they could make large language models behave in unexpected ways. This is concerning because these models are widely used for tasks like answering questions and generating text. |
Keywords
» Artificial intelligence » Fine tuning » Natural language processing » Nlp
