Google Geminihas been the talk of the town since earlier this month when itgot rid of the Bard brandingwhile alsoreplacing Assistant on Android devices. But despite the promising start to the month, things appear to have fallen off a cliff for Google, particularly with regard to image generation. After mounting comments from users about Gemini’s tendency to generate inaccurate images of people in the past week or so, Google decided topause this capability until further noticethis week. The company is now explaining what went wrong and what it plans to do in the future.
5 best Google Gemini prompts that take full advantage of the generative AI
Ditch boring emails and tedious tasks; these Gemini prompts will transform how you work
Senior VP at Google’s Knowledge and Information division, Prabhakar Raghavan, penned ablog postdetailing the issue at hand while also acknowledging that Gemini’s image generation capabilities “missed the mark.” Here’s what Raghavan had to say on the reasons behind the trouble:
First, our tuning to ensure that Gemini showed a range of people failed to account for cases that should clearly not show a range. And second, over time, the model became way more cautious than we intended and refused to answer certain prompts entirely — wrongly interpreting some very anodyne prompts as sensitive.
A combination of these two critical issues, Raghavan says, led to Gemini overcompensating in some areas while being overly conservative in other requests, resulting in “embarrassing and wrong” imagery. While the company does go into great detail explaining the inadequacies of Gemini — more specifically,Imagen 2, which is used by the chatbot for image generation — there’s no explicit mention of when this capability would be back on the chatbot.
For now, the blog post says Google will work on improving image generation with people “significantly” before bringing it back to Gemini. So it’s up to anybody’s guess as to how long that will take, but it’s safe to say Google Gemini won’t be able to generate images of people anytime soon. To the company’s credit, it has always maintained that LLMs (large language models) can get some things wrong, referring to what’s broadly known in the industry ashallucinations. In closing, Raghavan says that Google will do its best to deliver AI “safely and responsibly” to its users.
Gemini can still generate images of animals, but not when people are involved
Not all types of image generation features have left Gemini, though, with users still being able to generate photos of animals and inanimate objects. But you can’t pull up images of people even with a generic query like “generate an image of a person walking a dog.” But if your request has no people in it, like the example above, Gemini is more than happy to generate some decent images.
