Gemini, the new name for the Bard chatbot, generates images based on user descriptions in a similar way to OpenAI's ChatGPT. Like other models, Gemini is trained to not respond to dangerous or hateful commands and to incorporate diversity into its output.
However, some users have criticized this diversity for going too far, creating inaccurate images of people and historical figures. For example, the first US President George Washington was portrayed by Gemini as a person of color.
Other examples include an AI image of a Southeast Asian woman dressed as a pope, when in fact all 266 popes throughout history have been white men.
 |
Google pauses Gemini AI image generation feature |
Gemini even created "diverse" images of Nazi-era German soldiers, including an Asian woman and a black man dressed in 1943 military uniform.
Because Google has not released the parameters that govern the Gemini chatbot's behavior, it's difficult to get a clear explanation for why the software invents so many diverse versions of historical figures and events.
Gemini's issue comes days after OpenAI, the company that operates the popular ChatGPT, introduced a new AI tool called Sora that can generate videos based on users' text descriptions.
Google has previously acknowledged that the chatbot's erratic behavior needs to be fixed. Gemini may have prioritized forced diversity in creating images, leading to historically inaccurate depictions.
Google began offering image generation using its Gemini (formerly Bard) AI model in February, in a bid to compete with OpenAI and Microsoft’s Copilot. Like its rivals, the tool generates a collection of images based on user-generated text descriptions.
Source
Comment (0)