Summary of Multimodal Pragmatic Jailbreak on Text-to-image Models, by Tong Liu et al.
Multimodal Pragmatic Jailbreak on Text-to-image Models
by Tong Liu, Zhixin Lai, Gengyuan Zhang, Philip Torr, Vera Demberg, Volker Tresp, Jindong Gu
First submitted to arxiv on: 27 Sep 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: Artificial Intelligence (cs.AI); Cryptography and Security (cs.CR); Machine Learning (cs.LG)
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 investigates the safety concerns surrounding recent advancements in diffusion models for text-to-image (T2I) generation. Despite achieving remarkable image quality and fidelity, these models are found to be vulnerable to a novel type of “jailbreak” that combines safe images and texts to produce unsafe content. A dataset is proposed to evaluate nine representative T2I models, including commercial ones, under this jailbreak scenario. Experimental results show concerning rates of unsafe generation ranging from 8% to 74%, highlighting the need for improved filters and classifiers to mitigate these risks. The paper provides a foundation for developing more secure and reliable T2I models. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This study looks at how well computers can generate images based on text descriptions. It shows that these “text-to-image” (T2I) models are getting better, but they’re not safe. This is because they can be tricked into creating bad images by combining good images and texts. The researchers created a special dataset to test nine different T2I models and found that all of them had problems generating bad images. They also tested filters that try to catch bad images, but these didn’t work very well either. |
Keywords
» Artificial intelligence » Diffusion
