Google revealed a range of new artificial intelligence technologies and collaborations aimed at introducing advanced AI capabilities to enterprises. These announcements were part of Google’s Google Next conference held recently in San Francisco.
Additionally, Google’s subsidiary, Alphabet, publicly introduced an updated version of its proprietary AI chips and presented an enterprise-scale tool for watermarking and identifying AI-generated images. The company also showcased security tools and enhancements to its office suite.
This flurry of announcements reflects Google’s recent efforts to demonstrate its AI strategies, following Microsoft’s ambitious AI initiative that has been unfolding over the past year. However, Google Cloud’s CEO, Thomas Kurian, emphasised the importance of a deliberate approach for their enterprise clients, advising them to proceed methodically in treating AI development as a strategic endeavour, rather than succumbing to a fear of missing out (FOMO) on generative AI without clear purpose.
To enhance Google’s enterprise cloud services, 20 new AI models were added to its existing collection, bringing the total to 100. This expanded AI infrastructure includes agreements that grant Google Cloud customers access to Meta Platforms’ AI model LLaMa 2 and Anthropic’s Claude 2, a startup’s AI offering.
Google also revealed updated versions of its foundational AI infrastructure, enhancing performance and adding features. For instance, the new version of its text model, PaLM, now supports larger volumes of text input, facilitating the processing of lengthy documents such as legal briefs and books.
In the field of AI-generated image protection, Google introduced SynthID, a technology capable of invisibly watermarking digital image files to ensure their integrity, even after modifications or tampering.
Furthermore, Google unveiled AI updates for its suite of office software and security tools. One notable addition is an AI-powered tool designed to facilitate the migration of databases from Oracle to open-source alternatives, a historically challenging task.
Google AI chips
In anticipation of the full launch of its fifth-generation tensor processing unit (TPU), Google provided access to an optimised version of TPU, named TPU v5e, designed specifically for training large models and efficiently delivering content from these models. While not as potent as the upcoming flagship fifth-generation AI chip, TPU v5e can be grouped into “supercomputer” pods of 256 units, allowing cloud customers to tackle more complex computing tasks by connecting multiple pods together.
