Google explains what went wrong with its AI images
Kaitlyn Cimino / Android Authority TL;DR Google has now provided an explanation of what went wrong with Gemini after it generated inaccurate and offensive images of people. The tech giant says there were two issues that caused the AI to overcompensate. The generation of AI images of people would only be re-enabled after being significantly […]
Kaitlyn Cimino / Android Authority
TL;DR
- Google has now provided an explanation of what went wrong with Gemini after it generated inaccurate and offensive images of people.
- The tech giant says there were two issues that caused the AI to overcompensate.
- The generation of AI images of people would only be re-enabled after being significantly improved.
Google found itself in a sticky situation after Gemini was caught generating inaccurate and offensive images of people. The company has since disabled LLM’s ability to produce images of people. The company has now issued an apology, as well as an explanation of what happened.
In a blog post, the Mountain View-based company apologized for Gemini’s error, saying it was “clear that this feature missed the mark” and that it was “sorry, the feature did not work well” . According to Google, two things led to the creation of these images.
As we reported earlier, we thought it was possible that Gemini was overcorrecting something that was problematic with AI-generated images that reflect our racially diverse world. It seems that’s exactly what happened.
The company says the first problem has to do with how Gemini is tuned to ensure a wide range of people are represented in images. Google admits it failed to “account for cases that clearly shouldn’t show a range.”
The second problem comes from how Gemini chooses which prompts are considered sensitive. Google says the AI became more cautious than expected and refused to respond to some prompts.
For now, Google plans to keep people image generation on ice until significant improvements are made to the model.
