Google Vertex AI Welcomes Gemini 1.5 Pro: Unveiling New Features for Developers

Google Vertex AI has just opened to the powerful Gemini 1.5 Pro, the latest iteration of Google's large language model. This integration unlocks a treasure trove of new features to enhance your AI development workflow. Expect significant performance improvements, allowing you to tackle even more complex tasks more efficiently. Additionally, Gemini 1.5 Pro boasts a finer-tuned understanding of code and natural language, leading to more accurate and nuanced results in your projects. Google hasn't revealed all the specifics yet, but this collaboration between Vertex AI and Gemini 1.5 Pro promises to be a game-changer for developers working on the cutting edge of AI.

Google Cloud has announced several key upgrades to its Vertex AI platform, including new solid generative AI models and expanded corporate deployment features. If you're new to Google Vertex AI, it's a single platform consolidating the brand's several cloud-based AI technologies for corporate usage. You may utilise the platform to create ML models tailored to your organisation's needs and deploy them using pre-trained and custom tools.

What's New on Google Vertex AI Welcomes Gemini 1.5 Pro

At the forefront is a public preview of Gemini 1.5 Pro, which will provide developers with the world's most enormous context window of 1 million tokens. According to Google, such a broad window allows native reasoning over massive volumes of data relevant to each request, frequently removing the need for strategies such as fine-tuning or retrieval-augmented creation. One novel new feature is the ability to handle audio streams within Gemini 1.5 Pro using Vertex AI. This cross-modal capability enables seamless analysis across text, pictures, video, and audio sources, such as earnings calls.

However, Gemini 1.5 Pro is one of many models available. Google estimates it has around 130 models in all. The latest release also includes Anthropic's Claude 3 family of models and the lightweight CodeGemma models.

Grounding models using 'Enterprise Truth'

A major problem has been maintaining generative AI models in sync with reliable, up-to-date knowledge sources. Google addresses this by enabling Vertex AI models to base their replies directly on Google Search via a public preview feature. This is complemented by retrieval augmented generation (RAG), which grounds model outputs in an enterprise's data. Google refers to these grounding approaches as offering "Enterprise Truth" - a requirement for developing dependable, task-oriented AI agents.

Imagen's animation and editing capabilities

Google is demonstrating the capacity of its Imagen model to produce brief, 4-second live animated visuals from text prompts at 24 frames per second. While initially limited to 360x640 resolution, this innovative capability will significantly benefit marketing and content development teams. The widespread availability of advanced editing tools such as inpainting, outpainting, and undetectable digital watermarking is also enhancing Imagen 2's picture production capabilities. Inpainting removes undesired features, whereas outpainting widens the image boundaries to provide a larger view.

MLOps tools for timely management and assessment

Recognising the inherent problems of deploying big models, Google Cloud has enhanced Vertex AI's MLOps capabilities. The new Vertex AI Prompt Management solution solves pain points like prompt experimentation, migration, and tracking by including versioning, AI-generated ideas, and collaborative capabilities. The prompt management toolkit assists with tasks such as prompt versioning, variation comparison, human feedback gathering, and AI-powered prompt optimisation ideas. Evaluation services aid in assessing safety, factual correctness, and other essential performance parameters while models are iterated.