DeepMind CEO: Gemini and Veo AI Models Will Merge at Google

Google's Vision for Gemini and Veo Integration

During a recent podcast interview on Possible, featuring LinkedIn co-founder Reid Hoffman, Demis Hassabis, CEO of Google DeepMind, revealed Google’s intention to integrate its Gemini AI models with its Veo video generation models.

This integration is designed to enhance Gemini’s comprehension of the physical world. Hassabis explained that Gemini was intentionally developed as a multimodal foundation model from its inception.

The Pursuit of a Universal Digital Assistant

The core rationale behind this multimodal approach, according to Hassabis, is the ambition to create a truly versatile universal digital assistant. This assistant would be capable of providing practical assistance not just in the digital realm, but also in real-world scenarios.

The Rise of "Omni" Models in AI

The artificial intelligence sector is progressively shifting towards the development of “omni” models. These models are characterized by their ability to process and synthesize diverse media formats.

Google’s latest Gemini models demonstrate this capability by generating not only text and images, but also audio. Similarly, OpenAI’s standard ChatGPT model now includes image creation functionality, even supporting styles like those of Studio Ghibli.

Amazon is also actively involved, with announced plans to introduce an “any-to-any” model later this year.

Leveraging YouTube Data for Veo Training

Developing these comprehensive omni models necessitates substantial amounts of training data, encompassing images, videos, audio, and text. Hassabis indicated that the video data utilized for Veo is largely sourced from YouTube, a platform owned by Google.

“By analyzing a vast quantity of YouTube videos, Veo 2 is able to learn and understand the fundamental principles of physics governing the world,” Hassabis stated.

YouTube Content and AI Model Training

Google previously communicated to TechCrunch that its models “may be” trained on “some” YouTube content, adhering to agreements with YouTube creators.

The company reportedly updated its terms of service last year, partly to facilitate access to more data for training its AI models.

This adjustment allows Google to utilize a wider range of content to improve the performance and capabilities of its artificial intelligence systems.