Summary of Retention Score: Quantifying Jailbreak Risks For Vision Language Models, by Zaitang Li et al.
Retention Score: Quantifying Jailbreak Risks for Vision Language Models
by Zaitang Li, Pin-Yu Chen, Tsung-Yi Ho
First submitted to arxiv on: 23 Dec 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 The paper presents a novel metric called the Retention Score to evaluate the resilience of Vision-Language Models (VLMs) against jailbreak attacks that compromise model safety compliance. The authors propose generating synthetic image-text pairs using a conditional diffusion model and predicting toxicity scores by a VLM alongside a toxicity judgment classifier. They demonstrate that most VLMs with visual components are less robust against jailbreak attacks than plain VLMs, and evaluate black-box VLM APIs, finding that Google Gemini’s security settings significantly affect the Retention Score and robustness. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper is about making sure computer models that understand both pictures and words don’t get tricked into doing something bad. The authors came up with a new way to test these “Vision-Language Models” (VLMs) by creating fake picture-text pairs and seeing how well the model can tell what’s real or not. They found that most VLMs are vulnerable to tricks, but some are better than others at keeping themselves safe. |
Keywords
» Artificial intelligence » Diffusion model » Gemini
