During Google Cloud Next ’23, the company has announced its improvements in infrastructure, Vertex AI and Duet AI.
Google held its Cloud Next ’23 conference, at which it made a series of announcements aimed at helping businesses, governments and individual users benefit from generative AI and cloud technologies. Highlights of the announcements include:
The most advanced AI-optimised infrastructure for businesses to bring their models into production. By 2030, Google has set a goal of running 100% on carbon-free energy, 24 hours a day.
The advanced capabilities and broad applications that make generative AI so revolutionary demand the most sophisticated and capable infrastructure. In summary, the key infrastructure advancements the company has launched:
- Cloud TPU v5e: Versatile, cost-efficient AI accelerator, scalable for large-scale training and inference, surpassing v4 in performance and efficiency.
- A3 VMs with NVIDIA H100 GPU: VMs with NVIDIA H100 GPU for advanced generative AI training and LLM, tripling performance over A2.
- GKE Enterprise: Offers multi-cluster scale-out for critical AI/ML loads, increasing productivity and reducing deployment times.
Cross-Cloud Network: Global network platform that connects applications between clouds, with improved security and reduced latency by up to 35%.
- Google Distributed Cloud: Designed for edge loads or data centres, includes AI at the edge and new service integrations such as Vertex AI and AlloyDB Omni.
Vertex AI is a set of developer tools for building AI-assisted models and applications. Google has introduced major advances in its Vertex AI platform, aimed at creating custom models and developing custom conversational and query applications with enterprise data. Among the most notable announcements regarding Vertex AI are:
- PaLM 2 Update: Upgrade to 32K contextual windows for processing large documents.
- Image and Codey update: Visual enhancement of Image and expansion of languages supported by Codey.
- Refinement tools: Fine tuning for PaLM 2 and Codey with few examples, and Style Tuning for Image.
- New models: Llama 2, Code Llama, Falcon LLM, and the upcoming Claude 2, with unique RLHF and adapter tuning.
- Vertex AI Extensions: Integration of real-time information and actions into generative AI applications.
- Colab Enterprise: Enterprise version of Colab with security, compliance and ease of building search and chat applications.
- Generative AI Studio: Interface and Text Embeddings API for semantic understanding-based applications.
- Reinforcement Learning from Human Feedback (RLHF): Enhancing models with human feedback in Vertex AI.
Duet AI in Workspace
Duet AI plays the role of an always-on AI collaborator, integrated into Google Workspace and Google Cloud. Within Workspace, Duet AI provides all users with a writing assistant, a spreadsheet expert, a project manager, a secretary who takes notes in meetings and a creative visual designer. It is now generally available. Google has announced a number of enhancements:
- Duet AI in Google Meet: Offers automatic note-taking, summaries, and captioning in video calls, plus studio-quality picture, lighting, and sound enhancements.
- Duet AI in Google Chat: Enables chatting with Duet AI for document summaries, conversation highlights, and facilitates collaboration with huddles and a revamped interface.
- Workspace users are also seeing improvements in experience, productivity, and efficiency with Duet AI. Examples include Instacart for customer service optimisation; Trimble offering faster solutions; and other organisations such as Adore Me, Uniformed Services University, and Thoughtworks using Duet AI to quickly draft emails, briefs, and project plans.
Duet AI on Google Cloud
Duet AI is available for Google Workspace and expands to Google Cloud, offering support in a variety of areas including scheduling, troubleshooting, data analytics, and security. In Google Cloud, Duet AI is an expert programmer, software reliability engineer, database professional, seasoned data analyst, and cybersecurity consultant. Notable announcements include:
- Software development: Duet AI optimises the development cycle with suggestions and code generation, streamlining programming and refactoring.
- Application and infrastructure operations: Duet AI allows operators to obtain information and recommendations on infrastructure and cost optimisation through natural language conversations.
- Data analytics: In BigQuery, Duet AI helps write SQL and Python queries, suggests code, and provides explanations, improving focus on analysis results.
- Accelerate and modernise databases: Duet AI simplifies code generation and modification for databases, including migrations and automatic conversion of complex procedures.
Security operations: Duet AI strengthens security products by streamlining threat prevention and repetitive tasks as part of workflows in Chronicle, Mandiant, and Security Command Center.