New Delhi: Google has officially recognized issues with its Gemini model’s AI image generation, particularly concerning specific prompts. The tech giant stated that users requesting various images related to a particular culture or historical period should receive accurate responses. However, this hasn’t been the case, with Google attributing the problems to its “tuning” measures.
Explanation Of Shortcomings
In a recent blog post, Google delved into the factors contributing to problems with its Gemini model’s AI image generation. The company highlighted two main factors. (Also Read: From Investment To Income: A Rs 5-7 Lakh Investment In This Business Idea Could Yield Rs 1.5 Lakh Monthly Returns)
Firstly, their tuning process, aimed at ensuring Gemini could display a diverse range of people, overlooked scenarios where a varied representation wasn’t appropriate. Secondly, over time, the model became overly cautious and started declining certain prompts altogether. (Also Read: User Receives Fake iPhone 15 From Amazon; Company Responds)
Temporary Pause On Image Generation
Google admitted that their recently launched news image generation feature for the Gemini conversational app, which included creating images of people, missed the mark.
Some generated images were inaccurate or even offensive. In response, Google temporarily paused the image generation of people in Gemini while they worked on an improved version.
The company emphasized that this outcome wasn’t their intention and reiterated their stance against deliberately creating inaccuracies, especially with historical content.
Actions To Address Issues
To resolve the problems, Google plans to subject Gemini’s AI image generation to more testing. However, the company mentioned that they can’t guarantee Gemini won’t make mistakes or produce embarrassing, wrong, or offensive results even after fixing the issues. Nonetheless, they promised to take action whenever problems arise.
Recommendation To Users
While the Gemini AI model undergoes improvements, Google suggests users utilize Search’s AI image generation, which gathers “fresh, high-quality information” from the web.
Temporary Halt On Generating Images Of People
Following a backlash over inaccurate results, Google temporarily suspended its Gemini AI model from generating images of people.
This decision came after users shared images created by the model, primarily featuring people of color, including scenes from history that only involved white people.