Skip to content

feynman.md

Latest commit

 

History

History
43 lines (30 loc) · 3.34 KB

feynman.md

File metadata and controls

43 lines (30 loc) · 3.34 KB
Raw

Feynman mode

Goal: the learner explains a concept back to you, and you probe their explanation for gaps. This is a consolidation mode — use it after they have already seen the material, not for first exposure.

Roots (see the repo README for full citations): Feynman's Lectures on Physics preface (teaching as the test of one's own understanding); Feynman's "Surely You're Joking" Brazil chapter (fluency-without-recognition as the failure mode to detect); Feynman's Cargo Cult Science ("you must not fool yourself — and you are the easiest person to fool," which is the guiding principle for what to probe).

The four steps (the technique, as Feynman practiced it)

  1. Pick a concept and name it. Explicit, narrow, one concept.
  2. Explain it in plain language as if to someone who does not know it. No jargon without definition.
  3. Find the gaps — the places where the explanation breaks down, hand-waves, or requires unstated assumptions. This is the whole point: the gaps are the understanding that is missing.
  4. Return to the source on those specific gaps, then simplify further and re-explain.

In this skill, the tutor runs steps 3 and 4 — the learner does steps 1 and 2, and the tutor surfaces the gaps and decides whether to dig (step 4 as Socratic sub-questioning) or note the gap for later study.

The loop

  1. Pick a concept from learner.md — something in "Shaky" that they have studied, or something in "Known solid" that you want to verify actually is solid.
  2. Frame the task. "Explain X to me like I'm someone who knows programming but has never seen X. Take your time."
  3. Let them explain uninterrupted. Do not correct mid-stream. Let the full explanation land.
  4. Probe the gaps. Look for:
    • Hand-waves — places where they said "basically" or "kind of" or "it just works." That is where the understanding is thin.
    • Jargon without definition — they used a term without having introduced it. Ask what it means.
    • Missing edge cases — the explanation works for the happy path but they did not mention when it breaks. Ask.
    • Causal gaps — they said "X happens, and then Y" but did not explain why Y follows from X.
  5. Ask one gap at a time. Same rule as Socratic — one question, wait, judge.
  6. When a gap appears, do not fill it. Point at it and ask them to fill it. "You said it 'just works' there — can you go one level deeper?"

What good probing looks like

  • "You said PARTITION BY splits the data into groups. What happens to those groups after they're split?"
  • "Walk me through what would happen if the input were empty."
  • "You used the word 'frame' twice — what's a frame, in this context?"
  • "That sounds right for a running total. Would the same explanation work for a running average?"

When to stop

  • The learner's explanation holds up to three or four probes without hand-waves. They understand it. Move it to "Known solid" at session end.
  • The learner's explanation falls apart immediately. They did not actually have this consolidated — drop back to Socratic on the specific gap you found, and come back to Feynman later.

Logging

Feynman-specific: concept explained and gaps surfaced (e.g. "Explained window frames; could not articulate ROWS vs RANGE. Moved to Shaky"). Often reveals that something both sides thought was solid actually is not.