OpenAI Introduces Tools for AI Content Discovery
Date: 2024-05-09 Author: Gabriel Deangelo Categories: IN WORLD
AI lab OpenAI has unveiled a new tool for detecting images created using the DALL-E 3 algorithm, as well as a watermark that protects content from unauthorized access.
Classifier with 98% accuracy
The AI image classifier developed by OpenAI has an accuracy of 98%, making it a reliable tool for detecting the use of the DALL-E 3 algorithm even on processed and compressed images.
When processing images, such as cropping, compression, and changing saturation, the classifier accuracy drops to 95–97%, and when changing color correction, it drops to 82%.
C2PA Standards Initiative
OpenAI has begun adding "content credentials" to the metadata of images created with DALL-E 3 as part of its support of the Content Provenance and Authentication (C2PA) Coalition standards, of which the company is a member along with Microsoft and Adobe.
OpenAI notes that the new tools require improvements and encourages users to provide feedback to evaluate their effectiveness and functionality.
Collaborating on AI Transparency
OpenAI's AI transparency and accountability initiatives align with other companies such as Microsoft, which is also sharing its progress in deploying responsible AI systems.
