Yuvion VL: New multimodal LLMs target AI safety with adversarial robustness

Researchers have introduced Yuvion VL, a new family of multimodal large language models specifically designed for content and AI safety applications. These models are built with adversarial robustness in mind, employing a novel Confuse-then-Contrast Fine-Tuning method to enhance their ability to distinguish between visually similar but safety-critical content. The accompanying Yuvion VL RiskEval benchmarks demonstrate that Yuvion VL-32B achieves state-of-the-art safety performance, outperforming both open-source and closed-source commercial models while retaining general capabilities. AI

IMPACT Introduces specialized multimodal models for AI safety, potentially improving the detection and mitigation of adversarial content.

RANK_REASON Publication of a new research paper detailing a novel multimodal foundation model and associated benchmarks. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CV →

• AI safety
• Confuse-then-Contrast Fine-Tuning
• multimodal large language models
• Yuvion VL
• Yuvion VL-32B
• Yuvion VL RiskEval

• model release
• safety
• paper

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →