Building Natural Language and LLM Pipelines

2

Diving Deep into Large Language Models

The previous chapter helped us establish a conceptual foundation for solving problems with LLMs through pipelines and the generative AI project life cycle.

In this chapter, we will deepen our understanding of LLMs, tracing their evolution from the foundational 2023 baseline to the specialized, agentic architectures of 2025. We will begin by reviewing the core principles that defined the previous generation of models and frameworks, including the transformer architecture and the initial definitions of RAG, prompting, and fine-tuning. We will then explore the great specialization of 2024-2025: the paradigm shift away from monolithic models and toward two new, dominant classes: hyper-efficient small language ...