Summary of Uniguard: Towards Universal Safety Guardrails For Jailbreak Attacks on Multimodal Large Language Models, by Sejoon Oh et al.
UniGuard: Towards Universal Safety Guardrails for Jailbreak Attacks on Multimodal Large Language Models
by Sejoon Oh, Yiqiao Jin, Megha Sharma, Donghyun Kim, Eric Ma, Gaurav Verma, Srijan Kumar
First submitted to arxiv on: 3 Nov 2024
Categories
- Main: Computation and Language (cs.CL)
- Secondary: Artificial Intelligence (cs.AI); 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 The proposed UniGuard is a novel multimodal safety guardrail designed to minimize the likelihood of generating harmful responses in large language models (LLMs). This guardrail jointly considers unimodal and cross-modal harmful signals, allowing it to be seamlessly applied to any input prompt during inference with minimal computational costs. The paper demonstrates the generalizability of UniGuard across multiple modalities, attack strategies, and state-of-the-art LLMs, including LLaVA, Gemini Pro, GPT-4o, MiniGPT-4, and InstructBLIP, while maintaining their overall vision-language understanding capabilities. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary UniGuard is a new way to keep large language models from saying bad things. It helps by looking at both the words and pictures together to stop harmful responses. This guardrail can be used with any model and doesn’t slow it down too much. The research shows that UniGuard works well across many different types of inputs, attack methods, and state-of-the-art models. |
Keywords
» Artificial intelligence » Gemini » Gpt » Inference » Language understanding » Likelihood » Prompt
