GenAI on Google Cloud

Chapter 1. The Challenge of Generative AI Application Development

Dear reader, we are excited to have you on this learning journey with us. The objective for this first chapter is to introduce you to the unique complexities of building and deploying applications powered by large language models (LLMs). We hope this chapter will prepare you for subsequent deeper discussions of LLM operations (LLMOps) and agent operations (AgentOps)—really, extensions of machine learning operations (MLOps), which includes the behind-the-scenes work and set of best practices for managing the entire lifecycle of AI and agentic systems, from building and testing to deploying and responsibly maintaining these powerful systems in production.

Overview of LLMs, Generative AI Agents, and Potential Applications to Business Tasks

Let’s start by defining some key characteristics of LLMs and their application to generative tasks such as content generation, summarization, recommendation, problem solving, and discovery.

To break it down simply, an LLM is:

Large

LLMs are trained on enormous datasets, often containing billions or even trillions of words. Many LLMs have “read” a significant portion of the internet and many, many books. The corpus for many of the LLMs you may use as a consumer have been trained on the equivalent of several large libraries. In the case of multimodal model training, this includes libraries of text, images, videos, and audio.

Language

The primary focus of LLMs is on understanding ...