Google DeepMind is a research lab focused on developing safe artificial intelligence systems that help science and society. Its mission is to solve intelligence, an endeavor that requires significant research investments and the development of transformative technologies to address some of the world's most challenging problems.
DeepMind has a range of technologies to handle different aspects of AI. Its most powerful generative video model is called Veo, and its Gemini family of models is designed to handle multimodality, combining and understanding text, code, images, audio and video. Gemini models come in different incarnations, including Ultra, Pro, Flash and Nano, each geared for different purposes and use cases.
Gemini models are designed from the ground up to handle multimodality, which means they can reason across different modalities, including text, code, images, audio and video. They can generate code from different inputs, create images and text, and even handle tasks that span multiple languages. They also have a long context window up to a million tokens, which means they can handle long documents, thousands of lines of code, hours of audio and video.
Project Astra is an AI assistant that uses the Gemini models to rapidly process multimodal input and respond at a conversational speed. That makes it more natural and conversational.
DeepMind also has a focus on responsibility and safety in its AI work. It has a framework to assess and mitigate potential risks from large-scale AI models in the future and to acknowledge the importance of tackling ethical issues that may arise.
To use these technologies in apps, developers can employ tools like Google AI Studio and Google Cloud Vertex AI. That lets developers create new products and services that take advantage of the abilities of DeepMind's AI models.
Pricing details aren't disclosed, but you can sign up to try the latest technology and see what's new at Google DeepMind. Its focus on solving hard scientific and engineering problems means its technology is potentially useful for a broad range of industries and uses.
Published on June 23, 2024
Analyzing Google DeepMind...