Google Cloud, the enterprise division of Google, held its Born Digital Summit in Bengaluru on Wednesday. During the summit, Thomas Kurian, CEO of Google Cloud, said that the company has been working on AI since 2004 and is currently building a global infrastructure. India has two Google Cloud regions in Delhi and Mumbai respectively. Google Cloud currently offers two major AI solutions: Vertex for developers and Duet for end users and businesses. The company manages roughly 34 per cent of the internet traffic globally. Thomas highlighted that Google Cloud can train an AI model twice as fast and at half the cost compared to the competition. "Google Cloud model garden currently has over 100 models of different sizes and allows users to choose the model they want," says Thomas Kurian. The garden includes dedicated AI models for text, chat, programming code, images, audio, and visuals. "We upgrade our models every two weeks and perform a generational upgrade every four months," he added. Currently, 24,000 developers at Google are using the Google Cloud AI-powered coding model, and more than 75 per cent of them accept almost all the instructions. Moderna, one of the creators of the COVID-19 vaccine, is currently using Duet AI to accelerate the simulation of medical drugs. This is achieved by converting medical experiments into mathematical models and then running them on large models at a much faster speed than conducting chemistry experiments. He also highlighted how Gmail on smartphones is powered by a smaller AI model that runs locally and offers super-low latency. Similarly, larger models will have general skills, but they will be slower, as they have to evaluate hundreds of millions of parameters. Through the Cortex AI platform, Google Cloud offers search and conversations, where a user can interact with a system using natural language to retrieve information or transact with it. Google Cloud is also working on limiting the hallucination effect on its generative AI with a feature called grounding that provides fresh results along with fact-checks, where it compares the results of the model with a well-known dataset to limit the hallucination effect. Vision watermarking is another prominent feature of Google Cloud AI, where users will be able to identify AI generative images even if they are not watermarked, as images generated using AI models cannot be copyrighted. As a result of these efforts, Google recently unveiled a new generation of the PaLM (Pathways Language Model) model, which can summarise entire books. Med-PaLM is another AI model that provides medicine-related results. Another newly introduced model can generate photo-realistic images. On the end-user side, the Google Cloud Duet-AI-powered Help Me Write feature on Gmail has been used by over a million users since March 2023. Google Sheets also gets an AI-powered analyst, which can generate formulas and run calculations. Google Slides can now generate photo-realistic images with text inputs.
