Making Your AI Results More Predictable

Course details

Explore the fundamentals of working with AI and large language models, focusing on responsible usage, the intricacies of writing effective user prompts, and how to ensure human oversight while implementing strong, reliable systems. Instructor Ronnie Sheers shows you practical ways to recognize and mitigate biases, hallucinations, and randomness in AI outputs, as well as how to leverage system instructions for improved interaction efficiency. Get started applying few-shot learning techniques to generalize data more effectively. Find out why keeping a human in the loop enhances decision-making and accountability while also ensuring ethical system design through user feedback and risk management. Along the way, learn how to use moderation components, fine-tuning models for specific tasks, semantic similarity searches for better information retrieval, and retrieval-augmented generation (RAG) systems for enhanced system performance.

Concepts

0. Introduction

01 - Embracing the unpredictable
02 - What you should know
03 - Warning - Use responsibly
04 - Language model overview
05 - Hallucinations, biases, and randomness

1. Enhancing User Prompts

06 - The value and limitation of processing user prompts
07 - Leveraging system instructions
08 - Few shot learning
09 - Challenge - Use system instructions
10 - Solution - Use system instructions

2. Human Oversight

11 - Keeping a human in the loop
12 - Designing for human oversight
13 - The value of user feedback
14 - Conveying risks to users

3. Implementing Robust Systems

15 - Using moderation components
16 - Fine-tuning models
17 - Semantic similarity search
18 - RAG - Value and drawbacks
19 - Chunking approaches for RAG
20 - Adding re-rank to RAG systems

Conclusion

21 - Staying up-to-date with evolving tech