Google addresses concerns over the inaccuracies produced by its Gemini AI tool in a recent blog post,
attributing the issue to tuning errors that resulted in "embarrassing and wrong" historical images.
leading to instances like racially diverse Nazis and Founding Fathers.
Prabhakar Raghavan, Google's senior vice president, acknowledges that the model's tuning failed to account for cases where a specific range of people should not be depicted,
resulting in it refusing to generate images of certain demographics when prompted.
The over-cautious nature of the model caused it to "overcompensate" and become "over-conservative,"
Raghavan emphasizes the company's commitment to ensuring Gemini provides accurate representations of different people and scenarios.
While Google apologizes for the malfunctioning feature,
acknowledging that errors are an ongoing challenge inherent in large language models like Gemini.
Google has temporarily disabled the image generation feature of Gemini and pledges to enhance its capabilities through rigorous testing and improvements,