Google announced several major upgrades to its Vertex AI platform at its annual Google Cloud Next conference, aiming to strengthen its position as a leading provider of enterprise-ready generative AI capabilities.
The enhancements span new models, improved tooling for customization and responsible AI, and capabilities to connect models to real-time data and actions. According to Google, adoption of Vertex AI's generative AI features has accelerated rapidly, with the number of customer accounts growing over 15x in the last quarter.
New Curated Models Added to Model Garden
Google is expanding the selection of curated third-party models available through Vertex AI's Model Garden. New additions include Meta's Llama 2, Anthropic's Claude 2, and Falcon from the Technology Innovation Institute.
Model Garden allows users to easily browse, select, and fine-tune models for their specific use case and data. The variety of models gives enterprises flexibility to choose models aligned with their needs, including open source options for transparency.
Upgrades to Google's First-Party Models
Google also announced advances to its own generative models PaLM, Codey, and Imagen. Upgrades aim to improve quality, capabilities, and customization:
- PaLM 2 offers higher quality, 32k token context windows (e.g. an 85 page document), and grounding with enterprise data
- Codey delivers better code generation with 25% quality gains in major languages
- Imagen has visual improvements and new features like image editing and captioning
One innovative feature that's likely to grab attention is SynthID, a digital watermarking for Imagen-generated images. Developed in partnership with Google DeepMind, this feature embeds an invisible, tamper-resistant watermark directly into the image pixels, allowing for authentication without affecting image quality. This technological leap positions Google Cloud as the first hyperscale cloud provider to offer this level of security in AI-generated images.
Connecting Models to Real-Time Data and Actions
On the practicality front, Vertex AI is rolling out Extensions and data connectors. Extensions are fully-managed developer tools that connect models to APIs for real-time data and actions. These tools can be integrated with popular databases like BigQuery, MongoDB, and DataStax, opening up limitless use cases ranging from digital assistants to automated workflows.
This allows models to incorporate real-time data rather than stale, frozen information. Extensions enable building applications like chatbots that can look up live inventory or security tools checking code against latest threats.
Clients are already lauding the benefits. GitLab, for example, is leveraging PaLM 2 and Codey to deliver AI-powered workflows, while Omnicom is using Imagen to create audience-driven customized images. Other clients, like Workiva and Connected Stories, praised Vertex AI's capabilities to improve operational efficiency and content personalization.
Overall, the Vertex AI enhancements demonstrate Google Cloud's commitment to providing accessible, enterprise-ready generative AI capabilities with choice, customization, and responsible AI safeguards. Google aims to accelerate adoption by meeting developers and data scientists where they are with intuitive tools.