Google Cloud Platform to Incorporate AI Models from Meta and Anthropic

Google, a subsidiary of Alphabet Inc., is incorporating artificial intelligence tools from Meta Platforms Inc. and Anthropic into its cloud platform. This move aims to integrate more generative AI capabilities into its products and establish itself as a comprehensive solution for cloud customers looking to leverage this technology.

Google’s cloud customers can use Meta’s Llama 2 large language model and artificial intelligence startup Anthropic’s Claude 2 chatbot to edit business data for their own applications and services. The change, announced Tuesday at Google’s Next ’23 event in San Francisco, is part of the company’s effort to position its platform as one where customers have the freedom to choose the AI model that best suits their needs, either by themselves or from one of its partners. According to the company, there are now more than 100 powerful AI models and tools available to Google Cloud customers.

The company also announced the wider availability of its Duet AI product to customers of its Workspace productivity software, with the public following later this year. Users can tap a creative AI assistant that responds to prompts to help create content in apps like Google Docs, Sheets, and Slides. Introduced in May, Duet AI can take notes during video calls, send meeting summaries and translate subtitles into 18 languages, Google said. With the new “attend for me” feature, users can send the tool to attend meetings for them, deliver messages and create a summary of the event. Google also said it has new partnerships with companies like GE Appliances and Fox Sports. which allows customers to use artificial intelligence to, for example, create custom recipes or watch a replay of a sporting event from Fox’s broadcast list.

And Google announced a deepening partnership with chipmaker Nvidia Corp. Google said its cloud offerings are expanding to make more use of Nvidia’s chips and products designed to speed up the training of large language models. Google touted its access to Nvidia’s H100 accelerators — a prized commodity in the age of artificial intelligence — and said it would let customers use the latest version of the chipmaker’s so-called supercomputer.

With the announcements, Google says it’s more willing than ever to work with other AI companies as it seeks to gain market share from its rivals. Google has declared its products and services as the best alternatives in artificial intelligence, highlighting its years of experience in the field. Although the company still lags behind Amazon.com Inc. and Microsoft Corp. in the cloud computing market, Google said the AI additions to its cloud catalog give the platform the broadest range of models.

“We are living in a whole new era of digital transformation fueled by gen AI,” Thomas Kurian, CEO of Google Cloud, said in a blog post timed to coincide with the announcements. “This technology is already improving the operation of companies and the interaction of people with each other.”

In addition to adding new AI models to its cloud catalog, Google said it is making improvements to its own AI models and tools. PaLM 2, Google’s large language model that it announced at its annual developer conference in May, is now available in 38 languages and is better able to analyze longer documents such as research papers, books and legal documents, the company said. Meanwhile, Google’s coding assistance AI model, called Codey, has been updated to improve performance. The company’s text-to-image app, Imagen, includes higher-quality images and newer features, such as style tuning, to help cloud customers better customize their images to match brand guidelines, the company said.

Amid growing concerns about how companies should deal with the wave of AI-generated content, Google Cloud unveiled a feature that embeds a watermark to show that images are created by AI. The feature, based on technology from Google DeepMind’s artificial intelligence lab, includes a watermark at the pixel level, meaning it’s difficult to change, the company said.

Google also mentioned its major cloud customers and partners in its Tuesday announcements. The company said more than half of venture-backed generative AI companies pay for Google’s cloud computing platform, including Anthropic, Character.ai and Cohere. The company’s industry-specific models are also gaining traction, it said. Its Med-PaLM 2 model, an AI model for medical purposes, now offers partnerships with healthcare companies such as Bayer Pharmaceuticals, HCA Healthcare Inc and Meditech, Google said. Its Sec-PaLM 2 model, designed for cybersecurity, is used by service providers such as Broadcom Inc. and Tenable, Google added.

The cloud unit also announced a commercial service based on Ampere Computing’s new AmpereOne chip, adding weight to the startup’s claim that it can become a rival to chipmakers Intel Corp. and Advanced Micro Devices Inc., the dominant processor suppliers in data centers. Ampere, which is backed by Oracle Corp., claims its chips are more energy efficient than competing offerings and better suited to the high-performance computing needed by cloud service providers.