Friday, Bloomberg reported that Adobe Firefly's AI image generator incorporated AI images from competitors into its training data – a case of AI learning from AI.
The report brought a new dimension to the Adobe claims of Firefly ethics. In public articles, Adobe differentiated Firefly from competitors like DALL·E 3, Stable Diffusion and Midjourney highlighting Firefly's “commercially safe” training data.
“Adobe has established AI ethics principles of accountability, responsibility and transparency,” Adobe has written in a post.
Firefly drew from licensed Adobe Stock images, the company explained, plus images in the public domain. Adobe even created one bonus compensation plan for the artists whose work went into Firefly's first release.
Some of those artists submitted images created by Midjourney and other rival AIs, and were compensated by Adobe for their contributions, according to the Bloomberg report.
The symbol of one ouroboros, or the snake eating its own tail, created by Adobe Firefly in response to the Entrepreneur's request: “A snake is eating its own tail in an infinity symbol against a background of mountains, trees and a cloudy sky.” Credit: Adobe Firefly
Although artists had to note that their work was generated with AI, they didn't have to tell Adobe which generator they used.
This means that, even if Firefly isn't actively mining the Internet without permission, AI image generators like Midjourney it can be. And Firefly can train on those Midjourney images.
Connected: JPMorgan CEO Jamie Dimon says AI could impact 'every job'
Multiple Adobe employees told Bloomberg of an internal clash over the ethics of training Adobe's AI with AI-generated images from rivals. Even in the early stages of Firefly's development, some employees disagreed with the company's decision to include AI imagery in Firefly's training data.
IN Adobe Stock, the database used to train Firefly, 57 million have been labeled AI-generated, according to Bloomberg. This is 14% of the total images in the database.
Adobe responded to the claims by stating that only 5% of the images used to train Firefly came from AI images created by other platforms.
“Every image submitted to Adobe Stock, including a very small subset of AI-generated images, goes through a rigorous moderation process to ensure it does not include IP, trademarks, distinctive characters or logos, or artist names reference,” a company spokesperson said. told Bloomberg.
Connected: OpenAI reportedly used more than a million hours of YouTube videos to train its latest AI model
The race to develop the next big artificial intelligence has intensified the pressure companies face to find new sources of training data. According to the report from New York TimesOpenAI may have trained its latest text-to-video AI generator, Sora, on YouTube videos, and Google may have done the same.
According to statesman. Canva has 10.26% of the global market, per the same source.
A Bloomberg report from earlier this week shows that Adobe has begun paying its network $2.62 to $7.25 per minute of recorded video for new data to use in its competitor Sora.
Connected: Authors are suing OpenAI because ChatGPT is too 'accurate'