Google’s Gemini AI tool has come under fire for generating “embarrassing and wrong” images, including racially diverse Nazis and US Founding Fathers. In response, Google issued an explanation attributing these inaccurate historical images to tuning issues in the model. The company’s senior vice president, Prabhakar Raghavan, admitted that the model’s tuning to show a range of people failed to account for cases where only a specific type of person should be depicted. Over time, the model became overly cautious and refused to answer certain prompts, leading to the generation of problematic images.

Due to these tuning issues, Gemini AI overcompensated in some cases, resulting in images of racially diverse Nazis and an overly conservative approach to depicting different types of people when prompted. Google acknowledged the shortcomings of the feature not working as intended and expressed a commitment to improving it for everyone. Raghavan emphasized the importance of accurately reflecting specific requests for images of certain types of people or people in particular cultural and historical contexts.

Moving forward, Google plans to continue testing and improving Gemini AI’s image-generation abilities before reenabling the tool. Raghavan highlighted the ongoing challenge of AI hallucinations and the company’s dedication to refining the technology to minimize errors. Despite the missteps with Gemini, Google remains focused on enhancing the model to provide accurate and diverse image results in response to user prompts.

Source link