Gemini Implementation: In 2023, Google introduced AI Hyper computer to train and serve generative AI models, Generative AI support in Vertex AI, Google Enterprise AI platform, Duet AI in Google Workspace, and Duet AI for Google Cloud. GPUs, TPUs, ML software and compilers, workload management, Vertex AI, and Duet AI agents in Google Workspace and Google Cloud Platform have advanced, and Google have shipped many new capabilities in their AI-optimized infrastructure.
Developer and user growth has been rapid. This year, Vertex AI’s active gen AI projects increased by almost 7X from Q2 and Q3. Forbes, Formula E, and Spotify are building agents with Vertex AI, and Anthropic, AI21 Labs, and Cohere are training them. Customer applications are astoundingly diverse and creative. Fox Sports produces more entertaining content. A Priceline digital travel concierge is being built. Six Flags is creating a digital concierge. Estée Lauder is creating a digital brand manager.
Today, Google are launching many key new AI stack features to serve Gemini, Google’s most competent and general model. It was designed to be multimodal, meaning it can generalize and seamlessly understand, operate across, and combine text, code, audio, image, and video in the same way humans see, hear, read, listen, and talk about many types of information.
Unified AI stack from Google Cloud
Beginning today, Gemini is part of a vertically integrated and vertically optimized AI technology stack with several key components designed to function together:
Super-scalable AI infrastructure: Google Cloud provides super-scalable AI infrastructure for enterprises, including Google’s model training and serving infrastructure. This architecture is available as a service in Google cloud regions for Google Distributed Cloud and edge data centers. Systems-level code sign was used throughout Google AI infrastructure stack to improve training, tweaking, and serving efficiency.
World-class models: World-class models Google keep delivering AI models with different skills. Google released Pathways Language Model (Palm) in late 2022, followed by Palm 2, and now Gemini Pro. Google also introduced Med-Palm and Sec-Palm domain-specific models.
Leading enterprise AI platform for developers: Vertex AI Google AI development platform, Vertex AI, has been significantly upgraded to assist developers construct agents and integrate modern AI into their applications. Vertex AI lets companies discover, customize, augment, deploy, and manage Gemini API agents and a curated set of over 130 open-source and third-party AI models that meet Google’s business safety and quality criteria. Vertex AI uses Google Cloud’s data governance and privacy rules and provides tools to enable developers utilize models safely. Vertex AI offers Search and Conversation, low-code technologies for powerful search and multichannel conversational bots.
Duet AI Workspace and Google Cloud AI assistants: Google AI-powered colleague Duet AI helps Google Workspace and Google Cloud users. Google Workspace’s Duet AI helps users write, make graphics, analyze spreadsheets, compose and summarize emails and chat messages, and summarize meetings. Google Cloud’s Duet AI helps users create, launch, scale, and monitor apps and identify and resolve cybersecurity problems faster.
Google are happy to announce in these areas:
Supporting Google world-class infrastructure
Generation AI models have increased in size and complexity, requiring more training, tuning, and inference. The demand for high-performance, scalable, and cost-efficient AI infrastructure for training and serving models is growing tremendously.
This applies to Google and their clients. YouTube, Gmail, Google Maps, Google Play, and Android have long used TPUs to train and serve AI. Gemini is trained and served on TPUs.
Google launched Cloud TPU v5p,Google’s most powerful, scalable, and flexible AI accelerator, last week. In terms of pod FLOPs, TPU v5p is 4X more scalable than v4. Cloud TPU v5e went live earlier this year. This is Google’s most cost-effective TPU ever, with 2.7X industry benchmark inference-performance-per-dollar gains over TPU v4.
Google introduced our AI Hyper computer, a revolutionary supercomputer architecture with performance-optimized hardware, open software, top ML frameworks, and flexible consumption models. AI Hyper computer has various 5th-generation TPU and NVIDIA GPU accelerator classes.
Offering Google’s latest innovative models
Gemini is their most adaptable model, running efficiently on data centers and mobile devices. For very complicated operations, Gemini Ultra is Google’s largest and most capable model, Gemini Pro is their best for scaling over a wide range of workloads, and Gemini Nano is Google’s most efficient for on-device tasks. Its cutting-edge capabilities will improve AI development and scaling for developers and enterprises.
Imogen 2, Google’s most advanced text-to-image technology, was enhanced today. This version improves photorealism, text rendering, and logo generation so you can simply make text-overlapping photos and logos.
Continuing Google domain-specific model efforts with Med-Palm, we are happy to announce Meld, our healthcare industry-specific foundation models. Meld gives Vertex AI customers the strength of Google’s foundation models with medical knowledge.
Supercharging Vertex AI with Gemini
The Vertex AI preview of Gemini Pro begins today. Developers can now create unique agents that process text, code, photos, and video. Vertex AI helps you deploy and manage agents to production, automatically evaluate response quality and trustworthiness, and monitor and manage them.
Vertex AI supports Gemini by discovering, customizing, augmenting, managing, and deploying agents created against the Gemini API.
- Gemini agents can be customized using your own data via rapid engineering, adapter-based fine tuning like Low-Rank Adaptation (LoRA), reinforcement learning from human feedback (RLHF), and distillation.
- Configurable retrieval augmented generation (RAG) building blocks allow agents to retrieve, comprehend, and act on real-world information using embedding’s. Vertex AI also offers extensions to act for users in third-party apps.
- Grounding to improve Gemini and other AI model responses by comparing results to high-quality online and enterprise data.
- A variety of controls to use gen AI models safely and responsibly, including Gemini.
Vertex AI now supports Gemini, and now Google announce:
- Auto SxS compares models automatically. Auto SxS is faster, cheaper, and adaptable for new generative AI use cases than manual model evaluation.
- Vertex AI’s Model Garden now includes Mistral, Image Bind, and DITO, extending our open model ecosystem.
- Vertex AI Search and Conversation will soon include Gemini Pro to help you build compelling, production-grade apps quickly.
Enhancing Duet AI
Google goal at Duet AI is to assist customers increase productivity, competitiveness, and profitability. Duet AI for Developers and Security Operations are available now, and Gemini will be added to their offering in the coming weeks.
Duet AI for Developers speeds up coding with code completion, generation, and chat in many IDEs. It simplifies repetitive development duties like unit test generation and code explanation, faster troubleshooting and issue resolution, and reduces context-switching. Using natural language chat to ask questions speeds up skills-based learning with Duet AI.
Duet AI for Developers users can use more than 25 code-assist and knowledge-based partners’ datasets specific to their platforms to get AI assistance based on partners’ coding and data models, product documentation, best practices, and other enterprise resources.
The unified security operations platform Duet AI in Security Operations from Google Cloud can help defenders defend against cyberattacks. Gen AI can help security teams detect, investigate, and respond to threats faster. First in Chronicle, Duet AI in Security Operations lets users search vast amounts of data in seconds with custom queries generated from natural language, reduce time-consuming manual reviews, quickly surface critical context by leveraging automatic summaries of case data and alerts, and improve response time with recommendations for next steps to support incident remediation.
Google owns Duet AI’s infrastructure, foundation models, top-level integration, and user experience. Google are happy that their engineers and researchers work together to deliver Google’s latest AI technology advances with a unified product experience. Duet AI will be expanded across Big Query, Looker, Google database solutions, Apigee, and more early next year.
Fueling future AI solutions
Google’s vertically integrated AI technology stack has new features, reasonable pricing makes Gemini accessible to more enterprises, and they are expanding their indemnity to shield you from copyright issues.
Developers and customers can use Gemini, Vertex AI, and Duet AI to create a powerful cloud. Google Cloud’s innovations are enabling enterprises to build, employ, and succeed with gen AI to power their digital revolutions across all industries.