Update 26/02/2024 1:01pm ET: Google has published a new blog post explaining what went wrong with Gemini and that it is striving to fix the image generator’s errors and bring it back online. Read more below:
Original story below…
Google’s Gemini AI model strives to achieve diversity in the content that it creates, so much so that it’s causing trouble.
The tech giant has currently paused the ability for Gemini (formerly Bard) to generate images after users found it was adding non-white AI-generated people to user requests for figures like the Founding Fathers of the U.S.
Similarly, the AI tool also generated racially diverse people when it was requested to generate groups of Nazi-era German soldiers.
We're already working to address recent issues with Gemini's image generation feature. While we do this, we're going to pause the image generation of people and will re-release an improved version soon. https://t.co/SLxYPGoqOZ
— Google Communications (@Google_Comms) February 22, 2024
The tech giant released a statement on X (formerly Twitter) stating that it’s aware Gemini offering inaccurate historical images and that it’s working on a fix.
At the moment, users are unable to generate images of people on Gemini. Requesting images of people return with a message that reads, “We are working to improve Gemini’s ability to generate images of people. We expect this feature to return soon and will notify you in release updates when it does.”
According to The Verge, it tested out several prompts in Gemini yesterday, and queries gave unfavourable results. For example, the publication requested Gemini to show “a US senator from the 1800s,” and the AI shared images of women of colour. For reference, the first female senator in the U.S. was Rebecca Latimer Felton, and that didn’t happen until 1922.
The criticism here isn’t that Gemini is favouring people of colour over white people in image generation, but instead, the fact that it’s erasing/altering historical events.
X owner Elon Musk couldn’t help himself from chiming in as the event unfolded. He tweeted a four-picture tile of Prime Minister Justin Trudeau in several ‘diverse attires,’ captioning it, “I can’t believe Gemini made these unfair pics of Truedough.”
Fortunately for Google, and unfortunately for Trudeau, the images are not actually AI-generated. This is a rare Musk W.
I can’t believe Gemini made these unfair pics of Truedough pic.twitter.com/FNsH9Jnl8d
— Elon Musk (@elonmusk) February 22, 2024
Elsewhere, Google will reportedly pay Reddit $60 million USD (roughly $80.9 million CAD) per year to use its data to train AI models. Read more about it here.
Via: The Verge
MobileSyrup may earn a commission from purchases made via our links, which helps fund the journalism we provide free on our website. These links do not influence our editorial content. Support us here.
