We're off for the holidays so this is a re-release of one of our most popular episodes. Kyle and Jess completely overhaul their most popular episode on prompt engineering, because GPT-5 isn't just faster than GPT-4 - it's fundamentally different. It follows instructions with \"surgical precision,\" handles 800 pages of context at once, and will get confused if you give it contradictory prompts that older models would just ignore. Not to mention there's been a whole host of other updates to prompt engineering that deserve attention too.
We dive deep into why role prompting (like \"act as a marketing expert\") is largely ineffective according to new research, introduce the game-changing 4C Framework that actually works with GPT-5's precision, and show you how few-shot prompting can boost accuracy from 0% to 90%.
Whether you're frustrated with generic AI responses, wondering why your old prompts don't work as well anymore, or ready to master the communication skills that'll give you a massive competitive advantage, this episode is your roadmap to prompt engineering mastery in 2025.
What You'll Learn:
05:00 Revisiting Prompt Engineering - Why It Matters
08:16 Prompt Engineering: How Is GPT-5 Different?
17:53 Updating Commonly Held Beliefs About Prompt Engineering
19:58 WTF is Shot Prompting and How Does It Help Write Better Prompts?
21:45 Why You Need to Prioritize Your Context in Prompt Engineering
22:36 Decomposition and Self-Criticism in Prompt Engineering
25:32 Introducing the 4Cs Framework (+P) of Prompt Engineering
34:23 Applying the 4C Framework in the Real World
56:09 Quick Wins for Effective Prompt Engineering: Updated