Today in AI: Co-Intelligence Book Report, Jagged Frontier Explained, 4 Rules for Using AI

Introduction to Co-Intelligence

• 💡 The book "Co-Intelligence" by Ethan Mollick is highly recommended for its focus on mental models for thinking about AI, rather than specific technologies.
• 🧠 Released after GPT-3, it offers relevant insights into working with modern AI, unlike many older books that quickly became outdated.

Understanding the Jagged Frontier

• 🎯 The concept of the "jagged frontier" explains why AI can be superhuman at some tasks (e.g., advanced math) but surprisingly poor at similar ones (e.g., counting letters).
• 💡 This uneven capability means AI is a connection machine working on patterns, not a counting or computing machine, which can lead to misconceptions about its overall usefulness.
• ⚠️ People often dismiss AI as "rubbish" because they try to use it for tasks it's not designed for, failing to understand its architectural limitations.

Models for Human-AI Collaboration

• 🤝 The speaker presents two mental models from the book for working with AI: the Centaur Model and the Cyborg Model.
• 🐎 The Centaur Model involves a clear division of labor, where humans handle strategic thinking and client relations, while AI manages data analysis and drafting.
• 🤖 The Cyborg Model represents a deeper integration, with AI acting as a constant collaborator woven into creative and analytical processes, like a writing partner.

Essential Rules for Using AI

• ✅ Always invite AI to the table for new tasks to test its capabilities and learn its strengths and weaknesses.
• 🔍 Remain a human in the loop to monitor outputs, check work, and apply judgment, as blindly trusting AI can lead to errors, as seen in the Deloitte scandal.
• 🎭 Treat AI like a person but define its role, using frameworks like RISEN (Role, Instruction, Steps, End Goal, Narrowing) to give it specific expectations (e.g., "expert copywriter").

Continuous Improvement of AI

• 🚀 The fourth rule is to assume this is the worst AI you will ever use, recognizing that models are constantly improving.
• 📈 Dismissing AI based on current limitations is a "dumb bet," as what seems impossible today will likely be trivial in six to twelve months.
• 🧠 This mindset encourages continuous learning and adaptation, as even AI co-creators feel overwhelmed by the rapid pace of development.