Astra, as a multimodal AI, claims to help in complex tasks including visualizing the world, identifying objects, locating lost objects, and many more.
Google, at its I/O developer conference, showcased Project Astra, a real-time AI assistant. Its purpose is to become a universal helper for users. Demis Hassabis, head of Google’s DeepMind, demonstrated how this AI assistant can effortlessly perform complex tasks like finding your glasses. As a multimodal AI, Astra claims to assist in complex tasks including visualizing the world, identifying objects, locating lost objects, and many more. During the Google I/O event, Hassabis demonstrated Astra’s capabilities in a demo video, wherein the powerful AI model was answering questions in real-time while scanning the surrounding environment through a camera. With this, questions can be asked by pointing out things, and Astra can provide information about that object. For example, it’s as if you opened the camera, drew a circle or arrow towards any object visible on the screen, and asked, ‘What is this called?’ and immediately received a response from the smartphone describing that object, along with all necessary information. This is just one example; Astra can answer many complex and humorous questions in real-time.
Hassabis emphasized during the I/O the importance of AI agents that not only communicate but also execute tasks on behalf of users. He believes that the future of AI lies not just in fancy technology but in practical use. He discussed AI agents who not only converse but also perform many tasks on your behalf. He envisions a future where there will be different types of agents, from simple assistants to more advanced ones, depending on your needs and situation.
He mentioned that Astra’s development was made possible by improvements to Google’s language model, Gemini 1.5 Pro. Over the past six months, the team worked on making Astra faster and more responsive. This involved not only refining the model but also ensuring smooth operations on a large scale. Astra is one of several announcements from Gemini at this year’s I/O. Google also highlighted the 1.5 Flash AI model with Gemini 1.5 Pro, which has been developed to perform tasks like summarization and captioning faster. Additionally, Veo has been announced, which can generate videos from text prompts. There’s also Gemini Nano, a model designed to be used locally on a device like your phone.