OpenAI, which notably produces the artificial intelligence (AI) software ChatGPT, has announced that it is working on a tool that would allow to classify the origin of the images generated by its DALL-E 3 image generator.
OpenAI CTO Mira Murati mentioned working on this technology during Tuesday’s Tech Live event hosted by The Wall Street Journal.
The tool – which is still being tested internally – would determine with almost 99% accuracy whether an image has been created by DALL-E 3, the technical director said on stage.
However, Murati did not provide any information about a release plan for the detection tool but promises to provide more information soon.
There are other tools
Google has already announced the implementation of new features designed to make it easier for Internet users to determine whether a photo was created by an AI. This includes the addition of an “About Image” button that provides contextual information about its creation, such as the date it was uploaded or whether it has already been identified as fake by the media. Press.
The Californian giant also labels the images generated with the AI software Midjourney or the Shutterstock image database.
OpenAI recently removed from ChatGPT a small “regenerate answer” notice at the end of its generated texts, which made it easier to recognize essays created by the AI and posted as-is.
According to the decryptors, the tool also reformulated certain sentences, for example as AI or as an AI language model, to adopt more banal formulas.
The company said in January 2023 when launching a recognition tool that it was impossible to reliably recognize all text written with AI.
With information from Mashable