Google made a splash at its recent Cloud Next conference with a slew of updates to Vertex AI, its machine learning platform. These updates focus on three key areas: expanding Large Language Model (LLM) capabilities, streamlining Machine Learning Operations (MLOps), and introducing a new agent builder tool.
Supercharged LLMs: More Context, Audio Processing, and Live Images
Vertex AI now boasts a public preview of the mighty Gemini 1.5 Pro LLM. This powerhouse model tackles massive datasets with its 1-million-token context support. Google claims this eliminates the need for fine-tuning or complex retrieval augmented generation (RAG) techniques for many enterprises.
But that’s not all. Gemini 1.5 Pro dives into the world of audio with the ability to process speech and audio streams from videos. This unlocks exciting possibilities for cross-modal analysis, combining insights from text, images, and audio. Plus, it offers transcription capabilities for searching video and audio content.
Imagen 2, another LLM family within Vertex AI, received its share of love too. Now it boasts photo editing skills and the ability to conjure 4-second video snippets or “live images” from your text prompts. While the live image feature is still under wraps, photo editing and digital watermarking are ready for prime time.
Vertex AI welcomes a new member to the LLM family – CodeGemma. This lightweight model from Google’s Gemma lineage is perfect for scenarios where efficiency is key.
Grounding Your Models in Reality with Google Search
Google understands that LLM training data can become outdated or lack crucial information for enterprise use cases. To address this, Vertex AI now allows grounding models in Google Search in addition to your own data. This integration significantly improves response accuracy by leveraging the vast knowledge base of Google Search.
MLOps Made Easy: Experiment, Evaluate, and Refine
Vertex AI now offers expanded MLOps features to streamline your machine learning workflow. Vertex AI Prompt Management takes center stage, empowering teams to experiment with prompts, migrate them seamlessly, and meticulously track them alongside parameters.
This feature provides a centralized library of prompts with versioning, rollback options, and AI-powered suggestions for optimal performance. It even lets you compare prompt iterations side-by-side and add notes for better collaboration.
On the evaluation front, Rapid Evaluation (currently in preview) helps assess model performance during the prompt design phase.
Data Residency Gets a Boost
Google is expanding data residency options for Vertex AI. Data at rest for Gemini, Imagen, and Embeddings API’s can now reside in 11 new regions, including Australia, Brazil, and India. This provides greater flexibility and control for enterprises with specific data governance requirements.
Introducing Vertex AI Agent Builder: No-Code Generative AI Magic
To stay ahead in the game, Google Cloud unleashed Vertex AI Agent Builder, a no-code offering built for crafting virtual agents powered by Google’s Gemini LLMs. This tool leverages Vertex AI Search, Google’s out-of-the-box RAG system, to significantly accelerate agent creation compared to traditional, time-consuming RAG techniques.
Vertex AI Agent Builder boasts a range of pre-built components to simplify development, management, and maintenance of even complex virtual agents. Additionally, developers can leverage built-in RAG APIs for quick grounding input checks. For intricate projects, vector search is available to construct custom embeddings-based RAG systems.
The no-code approach extends to Vertex AI extensions (reusable modules connecting LLMs to specific APIs/tools), Vertex AI functions (helping developers describe functions/APIs for Gemini to intelligently select the right ones), and data connectors (ingesting data from enterprise and third-party applications).
Gemini Meets Looker: Supercharge Your Business Intelligence
The magic of Gemini is now woven into Looker, Google’s business intelligence offering. This integration unlocks exciting features like conversational analytics, report and formula generation, LookML and visualization assistance, and automated Google Slide creation.
A Feast of Data Analytics Updates
The data analytics party doesn’t stop there. Google introduced a managed Apache Kafka for BigQuery and a continuous query preview for the same service, further bolstering Vertex AI’s data analytics capabilities.
In essence, Google’s Vertex AI update is a treasure trove of advancements for developers and enterprises alike. With powerful LLMs, streamlined MLOps, a no-code agent builder, and deeper integration with Looker, Vertex AI is poised to be a game-changer in the world of machine learning.