15 Foundation models and emerging search paradigms

 

This chapter covers

  • Retrieval augmented generation (RAG)
  • Generative search for results summarization and abstractive question answering
  • Integrating foundation models, prompt optimization, and evaluating model quality
  • Generating synthetic data for model training
  • Implementing multimodal and hybrid search
  • The future of AI-powered search

Large language models (LLMs), like the ones we’ve tested and fine-tuned in the last two chapters, have been front and center in the advances in AI-powered search in recent years. You’ve already seen some of the key ways search quality can be enhanced by these models, from improving query interpretation and document understanding by mapping content into embeddings for dense vector search, to helping extract answers to questions from within documents.

15.1 Understanding foundation models

15.1.1 What qualifies as a foundation model?

15.1.2 Training vs. fine-tuning vs. prompting

15.2 Generative search

15.2.1 Retrieval augmented generation

15.2.2 Results summarization using foundation models

15.2.3 Data generation using foundation models

15.2.4 Evaluating generative output

15.2.5 Constructing your own metric

15.2.6 Algorithmic prompt optimization

15.3 Multimodal search

15.3.1 Common modes for multimodal search

15.3.2 Implementing multimodal search

15.4 Other emerging AI-powered search paradigms

15.4.1 Conversational and contextual search

15.4.2 Agent-based search

15.5 Hybrid search

15.5.1 Reciprocal rank fusion

15.5.2 Other hybrid search algorithms