New models and tooling from Vertex AI enable enterprise

Since announcing Generative AI support on Vertex AI less than six months ago, we’ve been thrilled and humbled to see innovative use cases from customers of all kinds, from enterprises like GE Appliances, whose consumer app SmartHQ lets users generate custom recipes based on their kitchen food, to startup unicorns like Typeface, which helps organizations use AI for compelling brand storytelling. Vertex AI client accounts increased 15 times in the recent quarter, indicating significant demand.

At Google Cloud Next, we’re excited to announce several Vertex AI enhancements that enable our customers to easily experiment and build with foundation models, customize them with enterprise data, and smoothly integrate and deploy them into applications with built-in privacy, safety, and responsible AI:

New Model Garden models to fulfil our customer commitment to choice and diversity in an open environment.

Meta’s Llama 2 and Code Llama, the Technology Innovation Institute’s Falcon LLM, and Anthropic’s Claude 2 are new.These announcements provide Google Cloud selected first-party, open source, and third-party models.

Updates to numerous first-party foundation models deliver Google DeepMind knowledge to clients. This features PaLM enhancements with higher-quality outputs, a 32,000-token context window for bigger document analysis, and corporate data grounding.

Codey and Imagen, our code generation and chat models, now function better and produce better images.

New Imagen digital watermarking using Google DeepMind’s SynthID

New tools to help organizations maximize model value. Vertex AI Extensions get models real-time data and real-world actions, while data connections get data ingestion and read-only access across sources.

Over 100 big Model Garden models, including Llama 2 and Claude 2

Many clients start their gen AI adventure at Vertex AI’s Model Garden, which offers APIs to selected huge models. Developers and data scientists can use Model Garden to choose the right models for their use cases based on capabilities, size, customization, and more, giving them powerful models and the flexibility to tune and deploy them at scale.

TII’s Falcon and Meta’s Llama 2 provide our consumers more options, while Anthropic’s Claude 2 is pre-announcing today. Model Garden’s diversity allows organizations fit models to their needs, while open-source solutions like Llama 2 and Falcon provide complete transparency into a model’s weights and artifacts for compliance and auditing assistance.

We made it easy to tweak these models because an organization’s data is crucial to model success. Only our cloud provider supports Llama 2 with adapter tweaking and Reinforcement Learning with Human Feedback. Organizations may tune Llama 2 with their corporate data while maintaining complete control and ownership. Our recently introduced Colab Enterprise data science notebook lets customers modify these models.

PaLM 2, Codey, and Imagen model and tuning updates

While expanding our model ecosystem, we are investing in first-party models and tooling. Today we announce advances in:

• PaLM 2: PaLM now supports 38 languages and lets clients use their own company data or private corpus. Our latest version of PaLM 2 for text and chat offers 32,000-token context windows, enough to accommodate an 85-page document in a prompt, to allow lengthier question-answer conversations and summarize and analyze huge materials like research papers, novels, and legal briefs.
• Codey: We increased Codey’s code generation and code chat quality by 25% in main supported languages.
• Imagen: We upgraded the look and included picture editing, captioning, and visual questions and answers. We also announced experimental digital watermarking more on that below.

Beyond model updates, adapter tuning, now widely accessible for PaLM 2 for text, and Reinforcement Learning with Human Feedback (RLHF), presently in public preview, allow businesses modify models with their own data and instructions. We also introduced Style Tuning for Imagen to assist clients match their photos with their brand requirements with 10 images or fewer.

Customers already notice great value

• GitLab chief product officer David DeSanto stated, “We are thrilled to be partnering with Google Cloud on delivering AI-powered workflows throughout the software development lifecycle.” GitLab is using PaLM 2 foundation models, including the Codey model family, to give secure software developers new AI-powered experiences. Our AI-powered processes, GitLab Duo, help enterprises release safe software faster.
• visuals are powering Omni, Omnicom’s open operating system, which will allow 17,000+ trained and certified users to produce audience-driven personalized visuals in minutes. Imagen has helped scale image production and personalization. Art Schram, Omnicom Annalect Chief Product Officer, said integrating it into our platform expands audience-powered creative inspiration to a new scale. We’re adopting styles, fine tweaking, and data-driven suggestions. We look forward to responsibly providing our users with appropriate visual inspiration.”
• Workiva Chief Technology Officer David Haila said the platform uses generative AI technology to increase productivity and efficiency and provide insights that lead to better and quicker data-driven choices. PaLM 2 on Vertex AI helps our clients access the whole potential of generative AI, giving a rich user experience and allowing new capabilities everywhere in their workflow. The model upgrades guarantee that we continue to deliver dependable, integrated reporting solutions to revolutionize corporate reporting and create significant change globally.”
• “Google Cloud”’s AI and API support has transformed our workflow. Our application design combines foundation and proprietary ML models to overcome real-time content personalisation scalability issues using Google Cloud’s integrated AI environment.
• Integrating Imagen through Vertex API has helped our clients speed up content production, personalize more, and gain more detailed insights, which boosts performance, said Connected Stories co-founder Tommaso Vaccarella. Beyond the innovation, strict data protection procedures guarantee us and our clients that sensitive data is secured. Google Cloud gives us the power and speed to launch cutting-edge enterprise-ready solutions.”

Digital watermarking by DeepMind SynthID verifies Imagen-generated photos

We will collaborate across Google to accelerate safe AI practices wherever possible. We are testing digital watermarking on Vertex AI with Google DeepMind to allow our customers to validate Imagen, our text-to-image model,-generated pictures.

Customers require technologies to recognize fake content for image production. Metadata used to identify fake photos can degrade image quality. Vertex AI’s experimental digital watermarking enables Google Cloud the first hyperscale cloud provider to provide undetectable and tamper-resistant watermarks in AI-generated photos. Google DeepMind SynthID, a cutting-edge technology, embeds the digital watermark directly into the pixel picture, making it undetectable to the human eye and difficult to alter with without harming the image. 

Extensions connect to real-world data and trigger actions

Foundational models are strong, but they are frozen after training, therefore they may offer stale results. Fully controlled developer tools for extensions, Vertex AI Extensions link models to APIs for real-time data and actions.

Developers may construct or utilize pre-built extensions to popular corporate APIs using Extensions. Extensions let developers construct sophisticated advanced AI apps like digital assistants, search engines, and automated processes.

A developer may utilize pre-built HR database extensions and Vertex AI Search to create a chatbot that helps workers find benefit deadlines and changing travel restrictions in natural language. Code vulnerability analysis apps are another example. Extensions let developers consume internal codebases and monitor real-time security concerns. Use scenarios are nearly unlimited.

Vertex AI will offer pre-built extensions for BigQuery, AlloyDB, DataStax, MongoDB, and Redis. Vertex AI will allow developers interact with LangChain, authenticate with private and public APIs, and protect apps using Cloud’s corporate security, privacy, and compliance controls.

Meeting users where they are

Our AI portfolio includes numerous upgrades like these Vertex AI updates to meet developers at their level, regardless of machine learning skills.

The customer momentum GA Telesis and Vodafone using gen AI to increase operational efficiency, Priceline using Vertex AI for consumer interaction innovations, and many others has been exciting. As we improve our enterprise-grade AI services, we look forward to further client success.

Developers and data scientists love Model Garden and its tweaking tools start here. Our new Colab Enterprise is a must-have for data scientists experimenting with, constructing, and deploying models. Learn more here.

Vertex AI Search and Conversation can enable developers get started with typical basic AI use cases like chatbots and bespoke search engines without AI knowledge or coding.

Visit our website and documentation to start using Vertex AI. Visit Google’s AI Adoption Framework’s new AI Readiness Quick Check to learn more about managing gen AI.

News source