- Google DeepMind CEO may think OpenAI’s text-to-video generator, Sora, has an edge.
- Demis Hassabis told a colleague it’d be hard for Google to draw level with Sora, The Information reported.
- His comments come as Big Tech firms compete in an AI race to build rival products.
Demis Hassabis may think OpenAI’s text-to-video generator is more advanced than one made by his own company.
According to The Information, the Google DeepMind CEO recently told a colleague that it could be difficult for Google to draw level with OpenAI’s Sora.
Sora can generate videos up to a minute long that can consist of “complex scenes with multiple characters,” OpenAI said when it was unveiled. Demos of the Sora-generated videos went viral in February and seemed to impress many people.
Google has its own text-to-video generator, Lumiere, which it revealed in a research paper in January. It can generate five-second clips with a text prompt.
The company, which recently released its experimental AI model Gemini 1.5, said in the research paper it’s crucial to create tools that can detect “biases and malicious use cases” to ensure it’s safe.
That’s because there’s a “risk of misuse for creating fake or harmful content” with Lumiere, and it faced such an issue with Gemini’s image generator in February. Google paused the rollout of its image generator after it produced images of people of color in historically inaccurate contexts.
Hassabis’ take on Lumiere compared with Sora comes as Big Tech firms continue in an AI race with rival products. But securing a victory hinges on getting a sizable amount of data to feed their systems and push the technology forward.
OpenAI’s chief technology officer, Mira Murati, told The Wall Street Journal last month she didn’t know if Sora had been trained on YouTube videos.
YouTube CEO Neal Mohan then told Bloomberg last week he had no knowledge of whether OpenAI was using YouTube content to train its video generator. He added that doing so would be a “clear violation” of the platform’s terms of service.
The New York Times reported last week that OpenAI used YouTube videos to help improve its speech recognition tool Whisper, citing unnamed sources.
According to the report, OpenAI transcribed more than one million hours of YouTube videos and the transcribed text fed into its GPT-4 model.
Google also used YouTube to transcribe videos and train its AI models, The Times reported. Doing so could infringe copyright held by content creators.
Google DeepMind and OpenAI didn’t immediately respond to requests for comment from Business Insider, made outside normal working hours.
