What you need to know
- Google’s DeepMind CEO Demis Hassabis stated the company hopes to have Gemini’s AI image generation tool back “in the next couple of weeks.”
- The company once again stated the tool was not “working as intended,” as Gemini was prone to producing inaccurate images of historical figures and groups.
- Google disabled Gemini’s image generation on February 22, with the company stating on February 23 that its “tuning” is likely the cause behind the inaccuracies.
Google has announced that it plans to relaunch Gemini’s AI image generation software “within the next few weeks,” according to the New York Post. Regarding Gemini’s image gen inaccuracies, the company’s DeepMind CEO, Demis Hassabis, reiterated, “We have taken the feature offline while we fix that.” He added, “We are hoping to have that back online very shortly in the next couple of weeks, a few weeks.”
Once again, Hassabis informed reporters that Gemini’s AI image generation problems were not “working the way we intended” during a conference at MWC 2024.
However, Google may have another problem on its hands as users on social media started testing Gemini’s chatbot feature. Gemini reportedly refused to state which was worse: present-day memes or Adolf Hitler. This was said to be in response to Gemini looking over tweets by Elon Musk on X.
While the AI chatbot stated Musk had been criticized for being “insensitive and harmful,” the bot added, “it’s up to each individual to decide who they believe has had a more negative impact on society.”
This then spurred Nate Silver, former head of data and polling for FiveThirtyEight, to call the situation “appalling.” They then declared that Google should “shut down” Gemini due to its “scary” response.
The controversy stems from Google’s removal of Gemini’s AI image generation tool on February 22. Users asking the bot to create images of historical figures or groups were met with unusually diverse photos — given the nature of the requests. One prompt regarding the creation of a “1943 German Soldier” provided one user with an array of people of color or Asian descent.
Google posted on X that it was “already working” to address the problems users have witnessed and that a pause would be required.
Last Friday (Feb. 23), the company formally addressed the issue by discussing its likely causes. According to Google, there is a problem with its tuning in Gemini as the company attempted to avoid falling into traps such technology had done in the past. However, despite its efforts, Google stated Gemini hadn’t accounted for situations where a wide range of diversity isn’t appropriate.
Moreover, Gemini had started refusing to produce a set of images to certain requests. Google stated the AI had grown too “cautious” as it tried to “overcompensate” for its inaccuracies.