Definition

Fine-Tuning

Fine-tuning continues training a pre-trained model on your own examples so its weights adapt to a specific task, style, or vocabulary. The book's position is that most teams no longer need it: 200K-token context windows let you put documentation and examples straight into the prompt. It still earns its place for high-volume cost savings, regulated domains with specialized terminology, or output formats prompting cannot make consistent.

Fine-Tuning diagram from Intelligence at Scale
Diagram from chapter 7, The New Landscape

Explained in

Chapter 7: The New Landscape

The current state of AI. What's changed, what it means for builders.

Related terms

This is one term. The chapter is the argument.

Intelligence at Scale: 22 chapters, 65,000 words, 80-plus diagrams. Kindle, paperback and hardcover on Amazon.

Buy on Amazon.com
← All terms