Two Ideas For Humans Learning from LLMs

How to Think More Effectively

"Learning from AI how to think better"

Transformer architecture and agentic planning reveal powerful patterns for human cognition

Idea 1: Transformer Architecture & Multi-Pass Thinking

50% of the solution is knowing the problem. Transformers solve this with Query, Key, Value attention.

Analogy: Librarian + Book + Contents

The transformer attention mechanism teaches us to think in three passes.

Scaled Dot-Product Self-Attention

A = softmax( QKT √dk )V
What am I looking for?
Where should I look?
What content do I extract?

Human Application

• Pass 1: Locate - What's important? Where to focus?
• Pass 2: Identify - Who you are, what you need
• Pass 3: Extract - Organize and structure the content

Dialogic Approach to Knowledge

In therapy or self-reflection:

  • 50% is finding what's important to you
  • 50% is the "what" and "how" of contents

Key insight: Multiple passes help us overcome cognitive fixation and limited working memory - just like transformers overcome limited context windows.

Idea 2: Agentic Planning & Reflection (3x-10x Performance)

Agentic solutions use planning and reflection to dramatically improve performance.

They overcome limited context windows through external planning.

Humans with mediocre memory and ASD fixation patterns can learn the same strategies.

Planning Phase

  • Todo lists - External memory
  • Design docs - First pass analysis
  • Pre-meeting notes - Multiple passes
  • Written thoughts - Before action

Prevents cognitive fixation and distraction

Reflection Phase

  • Journaling - Process experiences
  • Final passes - Review important outputs
  • Learning from mistakes - Calibration
  • LLM judge pattern - Self-evaluation

Improves accuracy like separate LLM evaluators

Real-World Applications

Breaking the ASD-1 engineer pattern:

  • Start side projects without getting distracted
  • Focus on specific tasks without context switching
  • Complete projects by using external planning

Key insight: Todo lists may seem boring, but they're exciting when they enable you to solve problems you previously couldn't tackle.