Video summary

How to Use AI Without Letting It Think for You

Main summary

Key takeaways

Educational

Main ideas / lessons

  • LLMs can weaken human thinking if used too much. Multiple studies are cited showing reduced performance or memory/accuracy when AI is relied upon.
  • The concept of “intuition rust” is introduced: the more people use AI, the more their own intuition can erode.
  • Writing is framed as thinking. Because of this, the speaker avoids outsourcing the core thinking/writing work to an LLM.
  • A structured, three-phase workflow is presented for using LLMs without letting them think or draft for you:
    1. Use AI for idea exploration and research
    2. Do original drafting yourself
    3. Use AI only for grammar/spelling corrections
  • The approach is aligned with decision theory:
    • Diverge first (expand options)
    • Converge later (decide and produce your own conclusion/idea)

Methodology / workflow

Overall principle

  • Use LLMs consciously and strategically.
  • Avoid letting the LLM think for you or write the first draft.
  • Let your own thinking drive the final ideas.

Three writing phases

Phase 1: Pre-writing (LLM-assisted)

LLMs are used for two purposes:

  • Landscaping (market/overview research)

    • Example: for a pricing strategy document
    • Prompts include:
      • How competitors price
      • Differences in pricing across Europe vs. Asia vs. the US
      • How pricing strategies have evolved over time

  • Mind-expanding activities (ideation/exploration)

    • Prompt for additional possibilities before choosing a direction.
    • Example framing: “What else could this be?” (i.e., expand the “universe of ideas”)
    • Goal: expand possibilities before you start converging.

Phase 2: Drafting and editing (no LLM)

  • Write the full draft yourself from beginning to end.
  • Edit once yourself, without AI help.
  • Explicit rule: Never let an LLM write the first draft.

Phase 3: Final polish (LLM-assisted for mechanics only)

  • Paste your completed draft into the LLM.
  • Instruct it to provide grammar and spelling advice/edits only.
  • Constraint: preserve the ideas—only fix wording errors so the final document reflects your own thinking.

Decision-theory mapping (workflow ↔ thinking stages)

  • Use LLMs to support divergence.
  • Use your own work to complete convergence:
    • Narrow the options
    • Decide and finalize your own idea

Context / why this approach fits the speaker

  • The speaker contrasts their situation with roles where LLMs primarily increase throughput.
  • Their role emphasizes decision accuracy and strategy, not just producing large amounts of text.
  • Therefore, they limit LLM use to avoid outsourcing judgment.

Speakers / sources featured

  • Speaker: The host/creator of the “Code to Care” YouTube video series (unnamed in the subtitles).
  • Referenced studies / sources (by description, not named individually):
    • A study involving students using AI that affects homework accuracy (includes % figures).
    • an MIT study about students’ ability to quote their own AI-written or AI-assisted essays (includes % figures).
    • A healthcare industry example involving patient problem identification performance (includes % figures).

Referenced concept / terms

  • Referenced concept/field: Decision theory (divergent vs. convergent thinking).
  • Term introduced: “Intuition rust” (described as intuition weakening with increased AI use).

Original video