The Weird World of AI Prompting: Key Lessons for Lifelong Learners
Kudos to Ethan Mollick and his quest to teach us better prompting
Today, in a departure from my usual content, I want to share some insights from a thought leader I greatly admire in the realm of AI prompting and teaching. If you’re not following Ethan Mollick’s Substack newsletter then I encourage you to do so. He is a Professor of Management at Wharton renowned for his expertise in entrepreneurship and innovation and everything he writes applies to a wide audience.
He recently published a fascinating post on his Substack titled "Captain's log: the irreducible weirdness of prompting AIs." This piece is brimming with pragmatic observations about the art and science of crafting effective prompts for LLMs - a skill set that will be increasingly valuable for all of us as we navigate this new era of AI-assisted learning and discovery.
Professor Mollick has also launched a companion website called "More Useful Things," which serves as a free resource library featuring a collection of AI prompts and techniques mentioned in his newsletters. These are the skills that I believe will become increasingly crucial for all lifelong learners as we embrace this new frontier of AI-enhanced exploration and growth.
There is no one-size-fits-all "magic" prompt. Context, desired outputs, and even the specific LLM you use will heavily influence what works. Blindly using purported "magic words" is often ineffective.
Proven prompting techniques do exist.
Adding context (persona, audience, format)
Providing examples (few-shot learning)
Using step-by-step "Chain of Thought" approaches
Prompting has a huge impact. A well-designed prompt can make seemingly impossible tasks easy for the LLM you use. Mollick's new paper "Prompting Diverse Ideas" found the right ones to enable GPT-4 to generate diverse, high-quality ideas nearly on par with human brainstorming groups.
Mastering prompting is challenging. It requires hands-on experience, intuition about an LLM’s "personality," and adapting to model improvements over time. Yet most users can get great results without obsessing over them,simply by engaging the AI conversationally.
A knowledge gap is emerging between experienced and new AI users regarding what the technology can do. Inexperienced users' naive prompting leads to subpar results that underestimate AI's true potential.
For an in-depth look at Mollick's findings, I recommend reading his full post and paper. But the core message is clear - as lifelong learners, investing some time into understanding effective AI prompting will pay dividends in our ability to embrace these tools in the pursuit of curiosity and personal growth. Embrace the weirdness as part of the fun!
I appreciate his calm practical perspective on where AI is now, what you can reliably do with it, and how to relax in the process of enjoying the interaction with LLMs. I like to say AI tools and LLMs don't answer your problems but they can lead to insights to help solve them. For that, I’m grateful.
Now, go read Substack titled "Captain's log: the irreducible weirdness of prompting AIs”. And don’t be shy about signing up for Ethan Mollick’s newsletter!