Google expresses deep regret: Gemini image generation paused for rework

In a new announcement by Google, Senior Vice President Prabhakar Raghavan revealed that the Gemini conversational app (formerly known as Bard) has temporarily halted its image generation feature for people. This decision comes after user feedback highlighted inaccuracies and offensive content in some generated images.

Launched three weeks ago, the feature was built on the Imagen 2 AI model and aimed to provide diverse and accurate representations of people. However, it fell short, producing images that were not only incorrect but also potentially harmful.

Raghavan explained that the feature’s tuning failed to account for specific scenarios where a range of people should not be shown. Additionally, the model became overly cautious over time, refusing to respond to certain prompts, leading to embarrassing and wrong images.

As a result, the image generation of people in Gemini has been paused for improvement, with plans for extensive testing before reactivation. Raghavan emphasized that Gemini is a tool for creativity and productivity, and while it aims for factual responses, it may not always be reliable for current events or sensitive topics. Users are advised to rely on Google Search for accurate information on such matters.

The incident highlights the challenges in AI development, especially in generating sensitive content like images of people. Raghavan assured that efforts are ongoing to improve AI technology responsibly, acknowledging that mistakes are part of the learning process.