Google announces new generative AI lineup in advance of Microsoft’s rumored GPT-4 debut

This morning, Google announced a laundry list of new generative AI capabilities and features for developers, through a PaLM API and in Google Cloud, as well as new integrations for users of Google Workspace, including in Gmail and Google Docs.

The announcements come just a month after Google unveiled its search chatbot Bard and less than a week after Bloomberg reported that a new internal Google directive “requires generative AI to be incorporated into all of its biggest products within months.”

The news also appears in advance of Microsoft’s highly-anticipated virtual ‘Future of Work with AI‘ event this Thursday. Thanks to comments last week by Microsoft Germany CTO Andreas Braun, that event is rumored to include the release a multimodal GPT-4, as well as a ChatGPT upgrade for Microsoft 365 applications such as Word and Outlook. [UPDATE: OpenAI’s GPT-4 was released today in a surprise announcement]

During a virtual press briefing yesterday, Google Cloud CEO Thomas Kurian said that the AI announcements “represent the culmination” of many years of work, including bringing together Transformer technology advances in reinforcement learning and advances in parallelism and orchestrating large training workloads.

These are the key announcements:

PaLM API and MakerSuite for app development

Google introduced a new PaLM API, which will allow developers to build on Google’s “best” LLM models.

The API will come with a tool called MakerSuite, which lets developers build prototypes. According to Google, over time it will also include features for prompt engineering, synthetic data generation and custom-model tuning.

A private preview is offered to select developers today, Google said, while a waitlist will be announced soon.

Generative AI capabilities in Google Cloud

Vertex AI is Google Cloud’s end-to-end machine learning platform that helps data science teams fast-track their ML model development and deployment from feature engineering to model training, low latency inference with enterprise class governance and monitoring.

Now, developers can use Google’s foundation models in Google Cloud, initially to generate text and images. Over time, generating audio and video will become options. Google Cloud customers will be able to discover models, create and modify prompts, fine tune them with their own data and deploy applications.

Google Cloud also announced a generative AI App Builder, which “connects conversational AI flows with out-of-the-box search experiences and foundation models.”

Google Workspace generative AI features

Google pointed out that over 3 billion users are already enjoying AI features in Gmail and Google Docs, such as Smart Compose and auto-generated summaries.

The company announced that a “limited set of trusted testers” will test a new set of generative AI features in Gmail and Google Docs, including the ability to adjust the tone to be more playful or professional.

An ‘open’ Google AI ecosystem

Finally, Google also announced what it called the ‘most open and innovative AI ecosystem’ with a series of partnerships, programs and resources.

For example, companies building foundation models, like Cohere and Anthropic, are already using Google Cloud’s infrastructure, GPU and TPU clusters to help train LLMs. Now, AI21 Labs, Midjourney and Osmo are also partnering with Google.

A variety of AI solutions also launched new or expanded partnerships with Google Cloud, including Aible, Anyscale, Gretel, Labelbox, Snorkel AI and Weights & Biases.

In addition, several AI application providers were announced as part of a new Built with Google Cloud AI Initiative, including Bending Spoons, Faraday, Glean, Replit and Tabnine. And a group of consulting companies, including Accenture, Deloitte, BCG and McKinsey committed to growing the Google Cloud AI and generative AI advisory and implementation services and capabilities available to customers.