This video was recorded at Capital Factory on February 6, 2024. Speakers: Rakshak Talwar and Jason Stevens.
Apologies for the occasional shaky footage—the audio is clear throughout.

A memorable and insightful lecture on deep learning and prompt engineering. Huge respect to the speakers, researchers, production team, and supporting institutions for making this happen.

Types of Prompts – Quick Overview

• Informational: Explain how photosynthesis works.
• Creative: Write a short story about a robot who learns emotions.
• Instructional: Give me a step-by-step guide to create a blog.
• Formatting: Create a table comparing electric cars and gasoline cars.
• Code: Write a Python script to sort numbers.
• Image: Analyze the diagram in this photo.
• Role-Based: Act as a professional career coach. Help me write a resume.

Video Chapters

• 00:00:00 Introduction to Direct Preference Optimization (DPO)
• 00:11:54 Potential of Leading Models
• 00:18:29 Training Llama 3 with Inherited Resources
• 00:31:21 Reformulation via Direct Policy Optimization
• 00:38:11 Mechanism of DPO Training
• 00:44:55 Fine-Tuning on OpenAssistant Dataset
• 00:50:18 Risks of Self-Preference Bias
• 01:00:52 Prompt Engineering and Addressing Hallucinations