02_Making_It_Stay — how do you learn so that it lasts?¶
Level: 101 · for anyone who passed the unit tests and failed the final
The first chapter says how to tell whether you have learned something. This one says how to keep it, and why the three techniques with the strongest evidence in all of learning research, spaced retrieval, interleaving and making the foundations automatic, feel worse than the habits they replace. A fourth lesson is about the session itself: why staying focused on a stuck problem keeps it stuck, and what stepping away adds. Each lesson shows the arithmetic behind one of them.
| # | Lesson | The question it answers |
|---|---|---|
| 1 | Spaced retrieval | Why does testing yourself beat rereading, and why does keeping a fact for ten years cost only nine reviews? |
| 2 | Interleaving | Why does a blocked practice sheet score higher and prepare you worse for a mixed exam? |
| 3 | Cognitive load | Why is the same formula 19 items for a beginner and 1 for an expert, and what does practice actually build? |
| 4 | Focused and diffuse thinking | Why do a thousand more focused steps end where three did, and what does stepping away add? |
The through-line¶
Memory fades, fast at first and then slowly, and pulling a fact out strengthens it most when it had nearly slipped. Lesson 1 puts both into a toy model and shows that with every gap a fixed multiple of the one before, the number of reviews grows only with the logarithm of how long you want to remember. Lesson 2 counts what a practice sheet rehearses: a blocked sheet lets you skip choosing the method, and choosing is most of what a mixed exam tests, so the sheet that scores higher predicts the exam worse. Lesson 3 goes inside the head: working memory holds about four chunks, a chunk is whatever practice has made automatic, and so the load of a page depends on its reader. Lesson 4 is about the hour of work rather than the weeks after it: focused thinking takes small steps from where the first idea landed, so it climbs the nearest hill and stops at its top, and a thousand more steps end where three did; stepping away takes big steps, which find the right hill but not its top, so the two have to alternate. The first three are what Robert Bjork calls desirable difficulties: they cost more during practice and pay on the test, which is why students rate them least pleasant and why the evidence table puts them at the top.
Po polsku, w skrócie¶
Pierwszy rozdział uczył sprawdzać, czy coś umiemy; ten uczy, jak to zatrzymać. Trzy techniki o najmocniejszych dowodach w całych badaniach nad uczeniem się: powtórki rozłożone w czasie ze sprawdzaniem się z pamięci, przeplatanie typów zadań i doprowadzanie podstaw do automatyzmu. Wszystkie trzy są w trakcie nauki trudniejsze i mniej przyjemne niż nawyki, które zastępują, i właśnie dlatego działają: pamięć wzmacnia się najbardziej, gdy odtwarzamy coś, co prawie uciekło; egzamin sprawdza wybór metody, który arkusz jednego typu pozwala pominąć; a pamięć robocza mieści około czterech porcji, więc każda podstawa doprowadzona do automatyzmu zwalnia miejsce na nowe. Czwarta lekcja, według Oakley i Sejnowskiego, dotyczy samej godziny pracy: myślenie skupione robi małe kroki i zatrzymuje się na najbliższym wzgórzu, a tysiąc kolejnych kroków kończy tam, gdzie trzy; odejście od problemu robi duże kroki i znajduje właściwe wzgórze, ale nie jego szczyt, więc oba trzeba przeplatać.
A note on the code¶
The memory model in lesson 1 is a toy with the right shape, and the SM-2 schedule it runs is Anki's ancestor as published. Lesson 2's counts and volumes are exact. Lesson 3's chunk counts are exact for the reader it models; its costs for a phone and a conversation are made up for the example, and the page says so. Lesson 4's landscape is invented to have two hills; what is exact is where each search strategy stops on it.