Learns general patterns of language, code, and world knowledge through pretraining.
Parameter knowledge becomes outdated and cannot directly observe private business state.Chapter 07
LLMs And Modern AI Systems: Why do large models still need external systems?
Transformers let models learn general patterns from large-scale text, but real products are more than a single model call. Modern AI systems place the model inside an external structure of context, retrieval, tools, memory, and evaluation.
Why do large models still need external knowledge and tools?
System Map
The system boundary from generation to deployment
Temporarily places the current task, instructions, examples, and retrieved passages into the model input.
Context capacity is limited, and incorrect material can amplify errors.Recalls relevant evidence from external documents, knowledge bases, or search results.
Retrieval quality determines answer quality; missing evidence cannot be cited reliably.Lets the model call calculators, code, search, databases, or business systems to perform actions.
Tool calls require permissions, parameter validation, and failure handling.Stores preferences across sessions, previous tasks, and long-term project state.
Memory needs selection, update, and forgetting strategies or it will pollute the context.Uses testing, red teaming, permissions, and monitoring to assess system reliability.
Evaluation cannot eliminate risk; it can only expose risk earlier.Connections
External systems turn model capability into controlled processes
Base Model → Context Window
This connection shows that a modern LLM application is rarely a single generation. The model works among evidence, actions, and constraints.
Retrieval → Context Window
This connection shows that a modern LLM application is rarely a single generation. The model works among evidence, actions, and constraints.
Base Model → Tools
This connection shows that a modern LLM application is rarely a single generation. The model works among evidence, actions, and constraints.
Memory → Context Window
This connection shows that a modern LLM application is rarely a single generation. The model works among evidence, actions, and constraints.
Eval / Safety → Entire System
This connection shows that a modern LLM application is rarely a single generation. The model works among evidence, actions, and constraints.
Historical position
It connects Transformers, RAG, and agents
Earlier problemModel parameters are not the live world
Pretrained knowledge is powerful but becomes outdated, and it cannot directly read private documents, business databases, or tool results.
What it solvesPlace capability inside a system structure
Context windows, retrieval, tool use, memory, and evaluation turn a model into a composable software component.
Remaining problemThe system can still fail
Wrong retrieval, tool misuse, prompt injection, permission leaks, and evaluation blind spots still require engineering controls.
Simplification note
This is a system map, not a real orchestration engine
This chapter uses a static diagram to explain common LLM application components. It does not call a real model, database, tool API, or evaluation service. The component relationships are teaching abstractions that explain why modern AI applications need structure outside the model.
References