Google Claims It Fixed Gemini’s “Degenerate” People
September 2, 2024
History revision is a problem. It’s been a problem for…well…since the start of recorded history. The Internet and mass media are infamous for being incorrect about historical facts, but image generating AI, like Google’s Gemini, is even worse. Tech Crunch explains what Google did to correct its inaccurate algorithm: “Google Says It’s Fixed Gemini’s People-Generating Feature.”
Google released Gemini in early 2023, then over a year later paused the chatbot for being too “woke,”“politically incorrect,” and “historically inaccurate.” The worst of Gemini’s offending actions was when it (for example) was asked to depict a Roman legion as ethnically diverse which fit the woke DEI agenda, while when it was asked to make an equally ethnically diverse Zulu warrior army Gemini only returned brown-skinned people. The latter is historically accurate, because Google doesn’t want to offend western ethnic minorities and, of course, Europe (where light skinned pink people originate) was ethnically diverse centuries ago.
Everything was A OK, until someone invoked Godwin’s Law by asking Gemini to generate (degenerate [sic]) an image of Nazis. Gemini returned an ethnically diverse picture with all types of Nazis, not the historically accurate light-skinned Germans-native to Europe.
Google claims it fixed Gemini and it took way longer than planned. The people generative feature is only available to paid Gemini plans. How does Google plan to make its AI people less degenerative? Here’s how:
“According to the company, Imagen 3, the latest image-generating model built into Gemini, contains mitigations to make the people images Gemini produces more “fair.” For example, Imagen 3 was trained on AI-generated captions designed to ‘improve the variety and diversity of concepts associated with images in [its] training data,’ according to a technical paper shared with TechCrunch. And the model’s training data was filtered for “safety,” plus ‘review[ed] … with consideration to fairness issues,’ claims Google…;We’ve significantly reduced the potential for undesirable responses through extensive internal and external red-teaming testing, collaborating with independent experts to ensure ongoing improvement,” the spokesperson continued. ‘Our focus has been on rigorously testing people generation before turning it back on.’”
Google will eventually make it work and the company is smart to limit Gemini’s usage to paid subscriptions. Limiting the user pool means Google can better control the chatbot and (if need be) turn it off. It will work until bad actors learn how to abuse the chatbot again for their own sheets and giggles.
Whitney Grace, September 2, 2024
Comments
Got something to say?
-
- Subscribe to Beyond Search
Feature archive
News archive
-
