Google ‘Messed Up’ Image Generation For Gemini AI Launch: Sergey Brin
Google co-founder Sergey Brin made a rare public appearance at an artificial intelligence (AI) event last weekend in San Francisco to acknowledge that the company “messed up” in its Gemini AI launch and platform features after the tool generated historically inaccurate images.
“We definitely messed up on the image generation,” Brin told entrepreneurs at the gathering, according to video shared by an attendee. “It was mostly due to not thorough testing.”
He acknowledged that the mistakes were upsetting to a lot of people, and the images prompted many developers and others to completely test the models.
The text models have two effects. “If you deeply test any text model out there, it will say some pretty weird things are out there and feel far left,” he said. “Just to be fair, there’s definitely work to do with this model.”
For those who tried it during this past week, Brin said the model’s results should be 80% better.
Google paused its AI image-creation tool in Gemini after it redrew its own vision of history.
Search & Performance Insider has learned that engineers across the company are not happy with the changes made. Some have decided to opt out of the company’s rules and find work elsewhere. The engineers did not specifically cite the AI mistakes — just the overall culture of the company. It’s not clear how this will effect products in the future.
Brin also cited the most promising applications running on Gemini on 1.5 Pro, such as long context-related apps. Developers have been dropping a ton of code or video content into the model and it will find the bugs.
When asked if developers will every really reach the point when they really understand how the models works, or will they remain black boxes and trust in the model becomes an important factor, Brin said “you can learn to understand. When we train these [models], there are a thousand capabilities you could try.”
The training cost of models are high, which is something the industry will need to cope with, he said. If a model can save a large amount of time for companies, he thinks if you measure the savings on a human productivity level and it can save someone an hour a week, the cost will pay for itself.
(12)
