Рубрики NewsSoftwareTechnologies

OpenAI has created a deepfake detector that recognizes images generated by DALL-E 3 with 98% accuracy

Опубликовал
Катерина Даньшина

The company that developed ChatGPT and DALL-E created is a tool for recognizing images generated by artificial intelligence.

Currently, the classifier works only with results created by OpenAI’s own image generator DALL-E 3 — but with high accuracy.

During internal testing, the tool correctly identified «artificial» images in 98% of cases. It is reported that the classifier can also identify typical modifications with minimal impactThe program uses a variety of techniques, such as compression, cropping, and saturation changes.

Starting today, OpenAI is accepting applications for access to the classifier for the first group of testers, including research laboratories and science-oriented journalistic non-profit organizations, within the framework of Access programs for researchers.

OpenAI also plans to add watermarks to mark digital content (photos or audio) that «cannot be deleted», and will join the industry group C2PA, which already includes Google, Microsoft, and Adobe.

Against the backdrop of the US and global elections Abuse of artificial intelligence tools is on the rise — In April alone, for example, fake videos of two Bollywood actors criticizing Prime Minister Narendra Modi went viral online.

Disqus Comments Loading...