Optimizing Prompts for Production - A Guide to Reliable AI Systems

Learn how to optimize AI prompts for production-scale applications in this 39-minute conference talk from GOTO Chicago 2024. Discover a systematic approach to prompt engineering that goes beyond simple trial and error, focusing on building reliable AI systems that deliver consistent results at scale. Follow along as prompt engineering expert Mike Taylor breaks down five key principles of effective prompting: giving clear direction, specifying format, providing examples, evaluating quality, and dividing labor. Explore practical optimization techniques including A/B testing, DSPy optimizers, and fine-tuning, while understanding the business case for implementing prompting solutions. Perfect for developers and technical professionals who have experimented with AI automation but struggled to achieve consistent, production-ready results.

Syllabus

00:00 Intro
02:27 Agenda
04:25 What is prompt engineering?
11:14 5 principles of prompting
11:36 1. Give direction
13:42 2. Specify format
16:54 3. Provide examples
20:29 4. Evaluate quality
25:14 5. Divide labor
28:22 Optimization in production
31:45 A/B testing
33:09 DSPy optimizers
34:20 Fine-tuning
38:07 The business case for prompting
38:29 Outro