Achieving F1 score of 83% with visual moderation is key. Expect enhanced safety.
Foundations for Safe Generative Media: Runway's Commitment
Original: Runway Research | Foundations for Safe Generative Media
Importance: 新たな安全性ガードレールの導入は多くのユーザーに影響を及ぼすため。
Summary
Runway has developed guardrails for safety, fairness, and integrity in generative AI models to support creativity. They built a visual moderation system to detect and block bad actors generating harmful content. Their model achieves an F1 score of 83%, and they have policies for child safety. Emphasis is also placed on catering to users from diverse cultural backgrounds.
Key Points
- Development of safety guardrails
- Implementation of visual moderation system
- Achieved F1 score of 83%
- Policies to protect children
- Diversity-conscious model
View developer notes (APIs, breaking changes, migration)
Runway has built an in-house visual moderation system capable of generalizing across both AI-generated and real-world content. This system automatically detects and blocks actors attempting to generate inappropriate material. The model achieves an F1 score of 83%, recall of 88%, and a false-positive rate of 2.8%, outperforming third-party APIs. They also train their models to reduce bias related to gender and race in prompts for certain professions.
Source: https://runwayml.com/research/foundations-for-safe-generative-media
Outlet: Runway
This article is an AI-generated summary (OpenAI GPT-4o-mini) of publicly available information from Anthropic, OpenAI, Google, Meta, Mistral, DeepSeek, Sakana, and other vendors. The original source URL is always provided in accordance with fair-use citation requirements. Summaries are AI-generated and may contain mistranslations or misinterpretations. Always verify details with the original source.