Summary of Fractured-sorry-bench: Framework For Revealing Attacks in Conversational Turns Undermining Refusal Efficacy and Defenses Over Sorry-bench (automated Multi-shot Jailbreaks), by Aman Priyanshu and Supriti Vijay
FRACTURED-SORRY-Bench: Framework for Revealing Attacks in Conversational Turns Undermining Refusal Efficacy and Defenses over SORRY-Bench (Automated Multi-shot Jailbreaks)
by Aman Priyanshu, Supriti Vijay
First submitted to arxiv on: 28 Aug 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 This paper introduces FRACTURED-SORRY-Bench, a framework for evaluating the safety of Large Language Models (LLMs) against multi-turn conversational attacks. The proposed method generates adversarial prompts by breaking down harmful queries into seemingly innocuous sub-questions, achieving a maximum increase of +46.22% in Attack Success Rates (ASRs) across GPT-4, GPT-4o, GPT-4o-mini, and GPT-3.5-Turbo models compared to baseline methods. The framework is built upon the SORRY-Bench dataset, highlighting the need for more robust defenses against subtle, multi-turn attacks. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper shows how to test if big language models are safe from bad conversations. The researchers made a new way to make tricky questions by breaking them down into smaller parts that seem harmless. They tested this method on different types of language models and found it increased the success rate of these attacks by 46.22%. This means we need better ways to keep our language models safe. |
Keywords
» Artificial intelligence » Gpt
