SAN FRANCISCO (Diya TV) — In response to concerns over accuracy, Google has decided to temporarily halt Gemini’s image generation feature, a move initiated after users criticized the tool for producing inaccurate and offensive images. The decision was announced by Google CEO Sundar Pichai, who addressed the company’s AI mishaps and emphasized the need for improvement.
Three weeks ago, Google introduced a new image generation feature for Gemini, allowing users to create images of people. However, the feature quickly drew criticism for its inability to accurately depict requested images, leading to a wave of user feedback and disappointment.
In a memo to employees, Pichai acknowledged the shortcomings of the feature, labeling them as “problematic” and admitting that they had “offended our users and shown bias.” Pichai further emphasized Google’s commitment to providing accurate and unbiased information across all its products, including emerging AI technologies.
Gemini’s image generation feature, built on the Imagen 2 AI model, was intended to ensure diversity in the generated images while avoiding the creation of inappropriate content. However, two critical issues emerged that led to inaccurate and embarrassing results.
Firstly, the tuning designed to display a range of people failed to account for cases where a specific type of person was requested. Secondly, the model became overly cautious over time, refusing to respond to certain prompts, further exacerbating the problem.
In response to these issues, Google made the decision to temporarily pause the image generation feature of Gemini. The company aims to significantly improve the feature before reactivating it, with extensive testing planned to ensure accuracy and reliability.
Pichai stressed the importance of continuous improvement and learning from mistakes, noting that while AI technology holds immense potential, it is still in its early stages of development. He reassured users that Google is committed to rolling out AI technology safely and responsibly, prioritizing the creation of helpful products deserving of users’ trust.
