1 Improving LLM accuracy

This chapter covers

Introducing the Large Language Models
Defining the limitations of LLMs
Shortcomings of continuously finetuning a model
Introducing retrieval augmented generation (RAG)
Combining both structured and unstructured data to support all types of questions

Large Language Models (LLMs) have shown impressive abilities across a variety of domains, but they have significant limitations that affect their utility, particularly when tasked with generating accurate and up-to-date information. To address these issues, this chapter begins by discussing the capabilities and constraints of LLMs, providing the necessary foundation to understand why external mechanisms like retrieval-augmented generation (RAG) are essential. However, RAG has often been limited to unstructured data, ignoring the potential of structured data sources like Knowledge Graphs (KGs), which are central to this chapter and the broader goals of this book.

1.1 Introduction to Large Language Models

1.2 Limitations of LLMs

1.2.1 Knowledge cutoff problem

1.2.2 Outdated information

1.2.3 Pure hallucinations

1.2.4 Lack of private information

1.3 Overcoming the limitations of LLMs

1.3.1 Supervised finetuning

1.3.2 Retrieval-augmented generation

1.4 Knowledge graphs as the data storage for RAG applications

1.5 Summary

1.6 References