The Next-Generation LLM(Large Language Model) on Google Cloud

As the scale and complexity of information grow, problem-solving becomes increasingly challenging. As a result, we are entering the era of Large Language Models (LLM). In response to this, Google recently unveiled Gemini, the next-generation LLM, last month. Google has announced plans to showcase and apply the performance of Gemini, an artificial intelligence (AI) model that surpasses its previous state-of-the-art LLM, ‘PaLM’.

Gemini signifies a one-step upgrade in the journey towards building more useful AI models for everyone. Unlike other AI models, Gemini has been trained to recognize, understand, and integrate various types of information, including text, images, audio, video, and code. This mirrors the way humans perceive, listen, read, speak, and comprehend diverse types of information simultaneously.

Gemini is categorized into three models – Gemini Ultra, Pro, and Nano – based on performance and size. Gemini Ultra stands as the largest and most powerful model designed for highly complex tasks, Gemini Pro is the top model with scalability across a wide range of tasks, and Gemini Nano is the most efficient model for on-device operations. The cutting-edge capabilities of Gemini are poised to significantly enhance the way developers and business clients build and scale AI.

Through Gemini, the Vertex AI platform undergoes a substantial enhancement.
  • You can now utilize Gemini Pro as a preview version in Vertex AI. Leveraging Gemini Pro, developers can build new and differentiated generative AI applications that process information across text, language, code, images, and videos. Vertex AI provides features to deploy, manage, and automatically evaluate and monitor the quality and reliability of generative AI applications in production environments.


    Built on the Gemini API, Vertex AI offers capabilities to utilize, customize, enhance, manage, and deploy applications comprehensively using Gemini.


    Developers can perform the following tasks through Vertex AI:

  • You can find and use the desired model from Google’s curated list of over 130 selected AI models, including Gemini. This list encompasses Google’s own models that meet strict corporate safety and quality standards, as well as open-source and third-party AI models. Developers can seamlessly integrate these models into their applications with easy API access.
  • Using tuning tools, you can enhance the learning knowledge and customize the model with specific domain expertise or proprietary knowledge held by the enterprise by adjusting model weights as needed. Vertex AI provides various tuning techniques, including prompt design, Low Rank Adaptation (LoRA), distillation, and other adapter tuning methods. Additionally, Vertex AI supports improving models by collecting user feedback through Reinforcement Learning from Human Feedback (RLHF) methods.
  • Developers can augment the model using tools that allow adjustments tailored to specific contexts or use cases. Utilizing Vertex AI Extension programs and connectors, developers can connect Gemini to external APIs for performing transactions and other tasks, retrieve data from external sources, and call functions from codebases. Additionally, Vertex AI provides the capability for enterprises to generate responses from foundation models based on their own data sources, assisting in improving the accuracy and relevance of model responses. Vertex AI supports grounding for both structured and unstructured data owned by the enterprise, as well as grounding using Google search technology.
  • Utilizing tools that support easy deployment and management of built applications, you can manage and scale models in a production environment. For this purpose, Google introduces a new model evaluation approach, the ‘Automatic Side by Side (Auto SxS)’ on-demand automation tool, to validate and compare models. Auto SxS is not only faster and more cost-effective than manual model evaluation but also allows customization of the evaluation process to adapt to various task types and address new generative AI use cases.
  • You can build search and interactive agents in low-code and no-code environments. With Vertex AI, developers can build outstanding production-level AI agents through Gemini Pro in just a few hours or days, regardless of their level of machine learning expertise, instead of weeks or months. Soon, Gemini Pro will be offered as an option to enhance the search result summarization and answer generation capabilities of Vertex AI, improving the quality, accuracy, and grounding capabilities of search applications. Additionally, the preview feature of Gemini for interactive agents, serving as the foundation model for voice and chat interactions, will be provided to offer dynamic interactions with users based on advanced inferences.
  • Responsible AI tools in Vertex AI, such as safety filters and the contents moderation API, ensure that models do not generate inappropriate content. This enables developers to implement responsible innovation.
  • Data can be protected on Google Cloud through built-in data governance and data privacy control features. Customers have complete control over access to their data, and Google does not utilize customer data for model training without explicit customer consent. Vertex AI provides various mechanisms, such as Customer Managed Encryption Keys (CMEK) and VPC Service Controls, to allow customers to have full control over their data.

Cloocus is quickly acquiring expertise in the emerging field of generative AI through its skilled Data & AI professional center. If you need expert consultation regarding the adoption of generative AI, please click the button below to request professional consulting!

If you need consulting services for cloud-based data and artificial intelligence, please contact Cloocus!
Secured By miniOrange