Want smarter insights in your inbox? Sign up for our weekly newsletters to get only what matters to enterprise AI, data, and security leaders. Subscribe Now
Today, Google Cloud kicked off its Cloud Next conference at Mandalay Bay Convention Center in Las Vegas.
The event – as many expected – saw Google make some major AI-focused announcements across its portfolio of cloud products. Thomas Kurian, the CEO of Google Cloud, touched on a range of aspects in his keynote, starting from new AI hardware and expanded integration of Gemini to upgrades for the models at play.
As everything played out, Vertex AI, Google’s premier platform for building, training and deploying machine learning (ML) projects, also came up in the conversation. The company announced it is making the offering better with several innovations aimed at making it more suitable for developers with different needs.
Here’s a rundown of the biggest improvements planned for Vertex AI.
AI Scaling Hits Its Limits
Power caps, rising token costs, and inference delays are reshaping enterprise AI. Join our exclusive salon to discover how top teams are:
- Turning energy into a strategic advantage
- Architecting efficient inference for real throughput gains
- Unlocking competitive ROI with sustainable AI systems
Secure your spot to stay ahead: https://bit.ly/4mwGngO
1. Support for models with longer context windows, live image generation
Vertex AI allows users to work with more than 130 models, including Gemini, partner models like Claude 3 and popular open models like Gemma, Llama 2 and Mistral. Now, building on this work, the company said it is making Gemini 1.5 Pro available in public preview, giving users access to longer context windows of up to 1 million tokens as well as the ability to process audio streams, including speech and the audio portion of videos, for cross-modal analysis.
Beyond this, Vertex AI is getting an improved Imagen 2 with the ability to create animated images lasting four seconds as well as a fine-tuned version of Gemma, CodeGemma, for code generation and assistance. Imagen 2 will also bring advanced photo editing features, providing users with inpainting and outpainting tools, Google confirmed.
2. Grounding capabilities for better results
While the new models deliver enhanced capabilities, they may not always provide accurate information. This is where Google’s all-new Search-based grounding feature comes into play. According to the company, the capability, which is now in public preview, combines the output of foundation models with fresh, high-quality information from Search to improve the completeness and accuracy of responses.
In case search-based grounding does not work, the company is also adding the ability to leverage Retrieval Augmented Generation (RAG) and ground models with data from enterprise apps like Salesforce or Workday.
3. New MLOps tools to improve performance
As Vertex AI has so many models to choose from, teams can find it difficult to identify the best-performing option for their needs. To help with this, Google Cloud said it is expanding the platform’s MLOps tools with solutions for prompt management and evaluation.
The former creates a collaborative library of prompts to help teams compare prompt iterations side by side and decide how small changes in a prompt can impact model performance. It includes versioning, the option to restore old prompts and AI-generated suggestions to improve prompt performance. Meanwhile, the latter enables a way to compare the performance of 1st party, 3rd party and open-source models at a specific set of tasks, complete with metrics for dimensions like instruction following and fluency. There’s also a new AutoSXS feature that provides detailed insights into why one model’s output is better than the other.
4. No-code Vertex AI agent Builder
Google also announced Vertex AI Agent Builder, an offering that brings together foundation models, Google Search and various developer tools to help enterprises build and deploy generative AI agents for different use cases. The best part here is that Agent Builder caters to developers with varying skill levels. One can either use a Gemini-powered no-code console for building AI agents with natural language prompts or switch to open-source frameworks like LlangChain on Vertex AI.
It also supports grounding via search and proprietary enterprise data.
5. Expanded data residency
Finally, in light of growing data sovereignty and regulatory requirements, Google announced new regions for its data residency effort. According to the company, enterprises using its generative AI services – Gemini, Imagen and Embeddings APIs – on Vertex AI can choose to store their data at rest in any of these 11 new countries: Australia, Brazil, Finland, Hong Kong, India, Israel, Italy, Poland, Spain, Switzerland, and Taiwan.
These regions add to the 10 countries Google had announced earlier across North America, Europe, and Asia and give users more control over where their data is stored and how it is accessed. There’s also a new option to restrict machine learning to the U.S. or EU when using Gemini 1.0 Pro or Imagen models.
Google Cloud Next 2024 will be live from April 9 to April 11.
