what you need to know
- Google officially stated that there was a problem with Gemini’s AI image generation, leading the company to disable the tool on February 22.
- The company said it tweaked Gemini to make it more diverse and sidestep AI-generated issues, however, it went too far and made the robot “cautious.”
- Google said it had brought the image-generating part of Gemini back for internal testing, saying the AI was prone to hallucinations and could still produce “inaccurate” results.
Google has issued a formal follow-up on an issue in which its Gemini model generated inaccurate AI images for specific prompts.
According to Google, when users prompt a bot asking for a series of images about a specific culture or historical timeline, they “absolutely deserve a response that accurately reflects” their intent. However, that’s not the case, and Google says its “tweaking” measures are where things went wrong in the first place.
“We tweaked it to make sure it doesn’t fall into some of the pitfalls we’ve seen with image generation technology in the past,” such as abusive or explicit images, the post added. Unfortunately, as Google explains, their adjustments don’t take into account the following: Wide The range of diversity is inappropriate.
The second flaw is that Gemini becomes “cautious”, rejecting prompts while being very sensitive towards others. Both of these issues are said to be catalysts for Gemini’s mistakes, as it attempts to “overcompensate.”
The company said this was “not our original intention.” Google also objects to intentionally creating inaccurate information about historical content. To address these issues, Google will bring Gemini’s AI image generation back into beta to iron out the issues.
The post adds that Google can’t guarantee that Gemini won’t hallucinate or create “embarrassing, inaccurate, or objectionable results,” even after the work is completed. However, the company claims that it will take appropriate action should issues arise.
Google, meanwhile, suggests users should use Search’s AI image-generating features because it uses a system that taps into “fresh, high-quality information” from around the web.
Yesterday (February 22) Google officially disabled Gemini’s ability to generate AI images based on user prompts. This was due to numerous user reports and criticism of the bot for inaccurately displaying historical figures and groups of people. While Google tried to encourage Gemini to create images of diverse people, it quickly became clear that this mentality didn’t apply to all prompts.
The company said yesterday that it “missed its target” and now we see the following corrective actions. Google hasn’t revealed when Gemini’s AI image generation will return.