Filling the internet with AI-created images will harm future AIs

The performance of image-generating AIs drops when trained using images previously created by other AIs, which can cause them problems given the proliferation of these images


November 21, 2022

A DALL-E photo of an astronaut on horseback

Photos produced by artificial intelligences could impair the performance of other AIs if used in training datasets


The popularity of text-to-image AIs could be their own downfall: if the images they produce proliferate too much, they could contaminate the datasets new models are trained on, hurting performance.

AI tools like DALL-E 2Midjourney and Stable Diffusion can create images of anything people ask for, and their images are increasingly being shared online.

Such AIs improve their results by training on datasets of billions of images and …

Leave a Reply

Your email address will not be published. Required fields are marked *