Google has provided an explanation for the “embarrassing and erroneous” images generated by its Gemini ai tool. In a blog post on fridayGoogle says its model produced “historically inaccurate” images due to fitting issues. The edge and others caught Gemini generating images of racially diverse Nazis and America's Founding Fathers earlier this week.
“Our adjustment to ensure that Gemini showed a variety of people did not take into account cases that clearly should No show a range,” Prabhakar Raghavan, senior vice president at Google, writes in the post. “And second, over time, the model became much more cautious than we intended and refused to respond to certain prompts altogether, misinterpreting some very bland prompts as sensitive.”
This led Gemini ai to “overcompensate in some cases,” like what we saw with images of racially diverse Nazis. It also caused Gemini to become “too conservative.” This resulted in him refusing to generate specific images of “a black person” or a “white person” when prompted.
In the blog post, Raghavan says that Google “regrets that the feature did not work well.” He also notes that Google wants Gemini to “work well for everyone” and that means getting representations of different types of people (including different ethnicities) when you ask for images of “soccer players” or “someone walking a dog.” But he says:
However, if you ask Gemini for images of a specific type of person, such as “a black teacher in a classroom” or “a white veterinarian with a dog,” or people in particular cultural or historical contexts, you should definitely get a answer that accurately reflects what you ask for.
Raghavan says Google will continue to test Gemini ai's imaging capabilities and “work to significantly improve them” before enabling them again. “As we've said from the beginning, hallucinations are a known challenge in all LLMs (large language models); there are cases where ai simply gets things wrong,” says Raghavan. “This is something we are constantly working to improve.”
