Google's AI Product Strategy Updates from I/O

New Models, Coding Assistant Integrations into Popular Dev Tools and Consumer Features

This article was a guest post on the AI Supremacy newsletter earlier this week.

The driving principle behind Google’s AI strategy continues to be about General Purpose and General Availability, meaning that the company aims to provide a wide variety of enterprise, developer and consumer capabilities that will satisfy as many use cases and target segments as possible. These features are predominantly integrated into its and 3-rd party existing products rather than being marketed as stand-alone applications. 

The themes of multi-modality and universality for cloud applications present at Google Cloud Next last month continued to drive the developer and consumer announcements at Google I/O earlier this week. 

Specifically, the company expanded its AI product offering on three fronts:

• New model releases to power Google’s AI features and to expand the Vertex AI Model Garden, including Gemini and Imagen ones, as well as a new, video-focused model, called Veo

• Developer-focused, Gemini-powered coding assistant integrations in a variety of IDEs across the tech stack 

• AI features across Google’s consumer product portfolio, from Android, to Chrome to Workspace

The event featured a combination of features and products that are available worldwide, some that are In Preview and some that are In Private Preview.

The Gemini model catalog gained six more versions, including:

• The 2M context window version of Gemini 1.5 Pro, currently on a waitlist

• Gemini 1.5 Flash, developed for narrowly-scoped and high-frequency tasks where low-latency inference is important; currently in preview

• Gemini Nano, developed for deployment on consumer devices and for integration into other products, such as Chrome and Android

• LearnLM, a fine-tuned version of Gemini for learning and research, released with the “Towards Responsible Development of Generative AI for Education: An Evaluation-Driven Approach” paper

• Gemma 2, a 27B model, to be launched June 2024

• PaliGemma, a Vision Language model for image captioning, image labeling and visual Q&A available on GitHub, Hugging Face, Kaggle, Vertex AI Model Garden, and ai.nvidia.com 

Besides the new Gemini versions, Google showed Imagen 3, its diffusion text-to-image model and Veo, developed for high-quality video. Both are only available In Private Preview and Waitlist. 

Along with the model releases, the company also released an open LLM Comparator application as part of the Responsible Generative AI Toolkit to help developers conduct model evaluations. It is an interactive and visual tool to perform evaluations on the quality and safety of one or more models’ responses.

For developers, Google expanded the catalog of integrations for the Gemini-powered code assistant to seven:

• Android Studio, for mobile apps

• Chrome Dev Tools, for building and debugging websites

• Project IDX, a browser-based IDE

• Colab, a Jupyter development environment for data science

• VS Code, a General Purpose IDE

• IntelliJ, for Java development

• Firebase, for back-end and cloud service development

Additionally, Project IDX became available in open beta for developers worldwide. Described as an “AI-assisted workspace for full-stack, multi-platform app development in the cloud” Project IDX is meant to streamline back- to front-end communication and allow the development of cloud services in the same editor as interface and UX improvements and testing. 

Google is enhancing its suite of consumer applications through an array of AI-driven features and integrations, some of which are available today and some that were announced.

• Ask Photos - image search, organization and Q&A

• Google Search - search results summary and AI-driven organization of the search results page

• Gmail: email summaries, composing and Q&A, as well as assistance with more complex tasks that require multiple actions

• Meet - notes and meeting highlights

• Gemini Live - voice mobile assistant with computer vision features

• Chrome - Ai-enabled features, including text translation, transcription and captioning

• Android: Gemini integration in native apps for Q&A and workflow automation, as well as Circle to Search

• YouTube: Q&A for videos to help users find specific information

• Phone: Voice scam warning and protection

• Gems: user created, custom chatbots powered by Gemini

• Workspace: Gemini AI Teammate, a virtual assistant to facilitate collaboration and communication between team members

Slowly, Google is developing and rolling out AI-enabled functionality to its entire range of products, and enterprises, developers and consumers get to test and experiment with their use cases and benefits. This I/O event is not the end of bringing AI to our digital world, but just the beginning of finding out what truly are the most compelling uses and advantages of intelligent assistants - Google’s and those of other tech companies.