DeepMind CEO Demis Hassabis Announces Future Integration of Google's Gemini and Veo AI Models

In a recent episode of the podcast Possible, co-hosted by LinkedIn co-founder Reid Hoffman, Google DeepMind CEO Demis Hassabis shared some exciting news about Google's plans. He revealed that Google is looking to merge its Gemini AI models with the Veo video-generating models. This fusion aims to enhance Gemini's grasp of the physical world, making it more adept at understanding real-life dynamics.

Hassabis emphasized that from the get-go, Gemini was designed to be multimodal. "We've always built Gemini, our foundation model, to be multimodal from the beginning," he explained. The motivation behind this approach? A vision for a universal digital assistant that can truly assist in everyday life. "An assistant that … actually helps you in the real world," Hassabis elaborated.

The AI industry is steadily progressing toward what you might call "omni" models—those capable of handling and synthesizing various types of media. Google's latest Gemini iterations, for instance, can produce not just text but also audio and images. Meanwhile, OpenAI's ChatGPT default model can whip up images on the spot, including delightful Studio Ghibli-style art. Amazon isn't far behind, with plans to roll out an "any-to-any" model later this year.

These omni models demand a hefty amount of training data—think images, videos, audio, and text. Hassabis hinted that Veo's video data primarily comes from YouTube, a treasure trove owned by Google. "Basically, by watching YouTube videos — a lot of YouTube videos — [Veo 2] can figure out, you know, the physics of the world," he noted.

Google had previously mentioned to TechCrunch that its models "may be" trained on "some" YouTube content, aligning with agreements made with YouTube creators. It's worth noting that last year, Google expanded its terms of service, partly to access more data for training its AI models.