what you need to know
- Google DeepMind CEO Demis Hassabis said the company hopes to restore Gemini’s artificial intelligence image generation tool “in the coming weeks.”
- The company again said the tool did not “work as expected” because Gemini could easily generate inaccurate images of historical figures and groups.
- Google disabled Gemini’s image generation feature on February 22, and the company said on February 23 that its “tweaks” may have caused the inaccuracies.
According to the New York Post, Google announced plans to relaunch Gemini’s artificial intelligence image generation software “in the coming weeks.” Regarding Gemini’s inaccurate image generation issue, the company’s DeepMind CEO Demis Hassabis reiterated, “We have taken the feature offline while we fix the issue.” He added: “We hope to soon in the next few weeks will be back online.”
At a conference at MWC 2024, Hassabis once again told reporters that Gemini’s AI image generation problem was not “working the way we expected.”
However, as social media users begin testing Gemini’s chatbot capabilities, Google may be facing another problem. Gemini reportedly refused to say which was worse: today’s memes or Adolf Hitler. This is said to be in response to Gemini checking out Elon Musk’s tweets on X.
While the AI chatbot said Musk has been criticized for being “insensitive and harmful,” the bot added, “Everyone can decide who they think has a greater negative impact on society.”
That prompted Nate Silver, former director of data and polling at FiveThirtyEight, to call the situation “shocking.” They then claimed that Google should “shut down” Gemini because of its “horrible” response.
The controversy stems from Google’s removal of Gemini’s artificial intelligence image generation tool on February 22. Users ask bots to create images of historical figures or groups, but the photos they get are unusually diverse given the nature of the request. A prompt for creating a “1943 German Soldier” provided one user with a list of people of color or Asian descent.
Google posted on X that it was “already working” to resolve the issues users were witnessing, but needed to pause.
On Friday (February 23), the company officially addressed the issue by discussing its possible causes. According to Google, its tweaks to Gemini are problematic as the company tries to avoid falling into the traps that such technologies have done in the past. However, despite its efforts, Google said Gemini did not account for situations where broad diversity was inappropriate.
In addition, Gemini has begun to refuse to produce a set of images based on certain requests. Google said the AI became too “cautious” as it tried to “overcompensate” for its inaccuracies.