How to Learn Anything Faster

Reread a chapter and it feels easier the second time. Your brain reads that ease as "I know this now." It's wrong — what you built is recognition ("looks familiar"), not recall ("I can produce it"). The exam, the interview, the real project all demand recall. This gap between feeling fluent and being able to produce is the single biggest money-waster in all of education.

✗ feels productive, mostly isn't:
   rereading · highlighting · watching another tutorial ·
   copying notes · "following along"

✓ feels awkward, actually works:
   closing the book and writing what you remember ·
   explaining aloud · doing problems cold · being tested

the rule: if it feels smooth, be suspicious.

The most replicated result in learning science: retrieving a memory strengthens it far more than re-exposing yourself to it. Testing isn't how you measure learning — testing is how learning happens. Every act of struggling to remember physically reinforces the pathway.

the practice swap, any subject:

after reading anything → close it → write 3 things you
remember → check → the gaps you found are the lesson

notes → questions:
✗ "Mitochondria produce ATP via cellular respiration"
✓ "What produces ATP, and how?"     (note = quiz card)

learned something Tuesday? → Thursday, try to explain it
to no notes. that struggle is the workout.

Memory decays on a curve — steep at first, flatter each time you review. The trick: review right before you'd forget, and each rescue flattens the curve further. Five reviews spread over a month beat twenty reviews crammed into one night, using a quarter of the time. Cramming works for tomorrow's exam and is gone by Friday; spacing is how things become permanent.

a schedule that works (and survives real life):

learn it        → day 0
review 1        → next day        (10 min)
review 2        → 3 days later    (5 min)
review 3        → 1 week later    (5 min)
review 4        → 1 month later   (3 min)

each review = active recall, not rereading.
tooling: Anki does this scheduling automatically —
worth it for languages, medicine, law, vocab-heavy fields.

Named for the physicist who could explain anything to anyone. Four steps: pick the concept, explain it in writing as if to a smart 12-year-old, notice exactly where you reach for jargon or hand-wave — that's where you don't understand it — go re-learn that piece, simplify again. Jargon is where understanding hides; being forced off it exposes what's underneath.

the tell-tale moment:

you: "...so the model uses attention to, um, attend to
     the important tokens..."
        ↑ you just used the word to define itself.
          THAT is the gap. that's where you go dig.

✓ works on paper, out loud, to a rubber duck, to an
  actual person (best: they interrupt with "why?")

Doing twenty of the same problem in a row (blocked practice) feels smooth and teaches little — after #3 you're on autopilot. Mixing problem types (interleaving) feels chaotic and slower, but forces the skill that actually matters: recognizing which tool this situation needs. Real life never announces the chapter number.

✗ blocked:      AAAA BBBB CCCC   (smooth, autopilot by A3)
✓ interleaved:  ABCA CABC BACB   (each item asks: which tool?)

in practice:
  math — mix this week's problems with the last 3 weeks'
  code — alternate: build, debug someone else's, read source
  language — mix new vocab into old sentence drills

Ten years of experience can be one year repeated ten times. What separates improvers from plateauers isn't hours — it's hours spent at the edge of ability, on the weakest sub-skill, with fast feedback. Comfortable repetition maintains; uncomfortable, targeted repetition grows.

the loop:
1. isolate ONE weak sub-skill      (not "get better at piano" —
                                    "left hand, bars 12-16")
2. work JUST past current ability  (fail rate ~20-40%)
3. get feedback fast               (teacher, tests, recording yourself)
4. fix, repeat, rotate weakness

The unifying principle behind everything above: effort during retrieval and processing is what writes memories deeply. Smooth input = shallow trace. This flips how you should read your own emotions while studying: frustration (within reason) is the feeling of it working; effortlessness is the feeling of it washing off.

same hour, different depth:

✗ watch solution video, nod along          (smooth, shallow)
✓ attempt 15 min → fail → THEN watch       (struggle primes it:
                                            now the solution sticks)

this is why: notes from memory beat transcribing,
generating a guess beats being told, and a hard first
attempt beats a perfect first demonstration.

The modern failure mode: finish tutorial, feel capable, start real project, blank completely, retreat to another tutorial. It's the illusion from section 01 industrialized — tutorials are recognition machines. The exit is a ratio and a rule: one unit of input, three units of output, and the project must not have a walkthrough.

the exit protocol:
1. pick a project slightly too hard, with NO tutorial
   (a real thing you want: a tracker, a bot, a tiny game)
2. get stuck → search for THAT specific problem only
3. finish ugly. working-and-ugly beats perfect-and-abandoned.
4. repeat. project #3 is where it clicks.

the searches change shape — that's the sign of progress:
"python tutorial" → "python read csv skip header" →
"why is my loop skipping items"     ← you're a developer now

The whole science, ready to use tonight.

# the meta-rule
smooth = suspicious. struggle (within reason) = it's working.

# after ANY learning session
close it → write 3 things from memory → gaps = tomorrow's list

# make it permanent
review at: 1 day → 3 days → 1 week → 1 month (recall, not reread)

# find your gaps
explain it to a 12-year-old. where you use jargon, you don't know.

# practice sessions
mix problem types · target the weakest sub-skill ·
work at ~30% fail rate · feedback fast

# skills (code, music, language)
1 unit input : 3 units output · projects without walkthroughs ·
finish ugly

# the 20-hour truth
zero → decent = ~20 deliberate hours. start counting.

One honest disclosure: this site is built on these principles — every SwiftGrasp page ends in a cheat sheet because retrieval beats rereading, and the audio exists so you can re-encounter ideas spaced over commutes. Now you know the machinery. Point it at anything — Python, mental math, whatever's next.