Vertex AI and Duet AI - Announcements from This Week's Google Cloud Next Event
Meta’s Llama 2 and Anthropic’s Claude 2 are now available within Vertex AI
At the Cloud Next event this week, Google announced new features and tools for its computing platform and AI infrastructure, including its foundation models, Vertex AI developer tools and the Duet AI companion application.
This post is a follow-up to my article on Google’s AI Product Strategy, hosted on the AI Supremacy newsletter, and it looks at how these announcements add to Google’s AI product portfolio, particularly Vertex AI and Duet AI.
Vertex AI Announcements Include New and Improved Models and Better Data Management Tools
Vertex AI is Google’s end-to-end Machine Learning platform, with tools and features for the entire value chain, from model selection to data preparation and application development. The recent advancements are around the creation of custom models and building custom Search and Conversation applications with enterprise data.
Specifically:
New and improved models available in the Model Garden:
PaLM 2 now has a 4 times greater input token length, up to 32k tokens, and is available in 38 additional languages
Codey, the coding companion based on PaLM 2, got a 25% improvement in quality compared to the previous version
Med-Palm, optimized for healthcare usage, is now in General Availability
Meta’s Llama 2 and Anthropic’s Claude 2 are now available within Vertex AI
Imagen, Google’s image foundation model, has new Style Tuning tools
New features for data management:
Vertex AI Search and Conversation are now generally available, enabling organizations to create Search and Chat applications with their enterprise data
Enterprise Grounding is a new service, now in Preview, which allows the use of a company’s data to fine-tune a foundation model and to add citations, in order to increase the factual accuracy of the AI application
The Extensions service, a set of fully-managed tools that help developers connect models to proprietary data or 3rd-party repositories now have integrations with Jira, Salesforce, Confluence.
Colab Enterprise is a new managed service for data scientists to work on AI workflows with access to the full spectrum of Vertex AI platform capabilities and integration with BigQuery
Tools:
The new Digital Watermarking feature offers a way to embed an invisible watermark into an image to make the identification of AI-generated art easier
These features complement the existing catalog within Vertex AI, which is composed of:
The Model Garden, which provides foundation and pre-trained proprietary and open source models
The Generative AI Studio to fine-tune models and build applications
MLOps tools, such as data labeling, pipelines and metadata management, to efficiently scale, manage, monitor and govern ML workloads
Integrations with the Google Cloud Platform and 3rd-party applications
The models available through Vertex AI span over 40 proprietary ones and over 60 that are open source. Per model type, they are 47 foundation models, 41 pre-trained, fine-tunable models and 35 off-the-shelf, task-specific models. Per data type, they are 35 language-based models, 74 for computer vision, 3 for video and 2 for tabular data.
Duet AI Is Now Available to All Google Workdspace Users and to Select Cloud Customers
Duet AI is a portfolio of AI features and tools that spans Google’s developer and consumer products, from the Cloud Platform, to Vertex and Workspace. Depending on the use cases for the particular product, it provides smart features to assist and speed up productivity work.
Duet AI’s recent announcements include:
General Availability in Workspace and expanded availability in Google Cloud.
New features in Google Meet include studio look, studio lighting and studio sound
In Google Chat, the tool has a refreshed user interface, new shortcuts and enhanced search
In Google Cloud, Duet AI will assist with code completion and code generation, as well as perform application and infrastructure operations through natural language
In BigQuery, the tool will provide contextual assistance for writing SQL queries and Python code and explain SQL statement in natural language
Besides these announcements, Duet AI is being developed to perform a variety of functions in the various Workspace applications, such as in:
AppSheet, Google’s NoCode data app development platform, where it will enable users to build workflow automation through natural language descriptions and conversation with the AI chatbot
Gmail, to assist with email drafting, including contextual assistance to fill in names and other relevant information
Docs, to help with text writing, draft generation and proofreading
Slides, to enable image generation through natural language
Sheets, to aid in data classification and analysis and the creation of task and project planning spreadsheets
Meet, to provide the ability to take notes
Alongside the Cloud Platform, Vertex AI and Duet AI are where Google is building its most powerful and versatile Machine Learning features and applications. Their performance in the market is a key driver for Google’s ability to capture more share of the growing cloud computing and Enterprise application markets. I’m looking forward to using more AI-enabled features in my productivity tools and to experience how these products make my work more interesting and efficient.