Adobe’s artificial intelligence model, Firefly, has been touted as a more ethical and commercially safe alternative to its competitors. However, recent reports have revealed that Adobe has been using AI-generated content from its rivals to train Firefly.
Firefly, Adobe’s flagship artificial intelligence product, was released in beta last March. It was initially trained on Adobe Stock, a database of 300 million stock photos and other visuals often licensed to marketers and media companies for ads and articles. However, to improve Firefly’s performance and make its output less commercial-looking, Adobe needed more photos of everyday items. As a result, Adobe turned to AI-generated content, including those from its rivals, to train Firefly.
This revelation contradicts Adobe’s claims that Firefly is a more ethical alternative. The company had previously criticized AI-art generators like Midjourney and Dall-E for indiscriminately swallowing data from across the internet, which led to intellectual property issues. Adobe had positioned Firefly as a safer alternative since it was mainly trained on Adobe Stock.
The company believes that the use of AI-generated content for training does not compromise the ethical stance of Firefly.
Read more at: www.bloomberg.com