text.extract_ideas

Goal

• You are an expert reader and critical thinker
• Print each step before executing it

Workflow

Read the Given Text

• Analyze the text below and extract the most thought-provoking, surprising, and intellectually interesting ideas

• Focus on:

  • Ideas that challenge common assumptions or conventional wisdom
  • Concepts that are counterintuitive, paradoxical, or unexpected
  • Insights with deep implications (philosophical, scientific, social, or psychological)
  • Particularly elegant, novel, or powerful ways of explaining something
  • Any hidden patterns, connections, or underlying themes

Extract Ideas

• For each idea create bullet points using following .claude/skills/markdown.rules.md and .claude/skills/text.rules.md

• Create an header 1 with a short summary

  # <ID>. Creative Destruction as Generative Force
  

• State the idea clearly and concisely

  **Idea**: ...
  

• Explain why it is interesting, surprising, or important

  **Why it is interesting**: ...
  

• (Optional) Add a short reflection or question that deepens the insight

  **Reflections**: ...
  

• E.g.,

  # 1. Pre-training as "Crappy Evolution"
  - **Idea**
    - ...
    - ...
  - **Why it's interesting**:
    - Pre-training LLMs on internet data creates a practical shortcut to
      biological evolution: encoding vast human knowledge into neural weights
    - Not true evolution (doesn't run on biology), but achieves similar
      bootstrapping in weeks instead of millions of years
    - Reframes AI development from "mimicking nature" to "using human cultural
      artifacts as evolutionary substitute"
    - Suggests evolution was finding an algorithm; pre-training is compressing
      knowledge
  - **Reflection**
    - If pre-training is evolution's shortcut, what breaks down in the analogy?
    - Is intelligence from knowledge compression equivalent to intelligence from
      algorithmic discovery?
  

• Prioritize quality over quantity: select only the most compelling ideas

Save the Output

• Save the output in <FILE>.ideas.md
• Run lint_text.py -i to format <FILE>.ideas.md

Verification

• [ ] Confirm <FILE>.ideas.md was created
• [ ] Confirm lint_text.py -i ran without errors
• [ ] Confirm each idea includes Idea, Why it is interesting, and (optional) Reflections