Google’s Remarkable Breakthrough in AI — Project Astra

Author(s): Sai Viswanth

Originally published on Towards AI.

Decode Project Astra Secret with new model updates.

Many big AI companies have started to focus on bringing Multi-Modal large Language models to the market.

OpenAI & Google released their flagship upgraded versions of their Multi-Model models in this year’s keynote GPT-4o & Gemini 1.5, respectively. Before the introduction of multi-modal models, separate models were trained to deal with one kind of modality. If you want to generate images, you need to run on a Model A, for audio on a Model B, text on Model C, and so on.

The idea is to have all inputs in one model:-

AudioVideosImagesText

The multi-modal models are kind of coming closer to replicating humans' perception of the outer world. Integrated with Robots, it can seemingly perform and enact like a human. Their strength comes from the hybrid architecture formed by the Transformers to effectively learn from long texts & Mixture of Experts (MoE) for conditional computation.

What is Project Astra & some of its interesting updates?The backbone of Project AstraHow did Google manage to speed up the model by at least 3 times than its competitors?Google Keynote Demo

Project Astra represents a significant achievement by Google DeepMind engineers, aiming to create a universal AI agent capable of handling any type of task. This ranges from simple… Read the full blog for free on Medium.

Join thousands of data leaders on the AI newsletter. Join over 80,000 subscribers and keep up to date with the latest developments in AI. From research to projects and ideas. If you are building an AI startup, an AI-related product, or a service, we invite you to consider becoming a sponsor.

Published via Towards AI