How to learn anything
This site is mostly about languages. This part is not. Learning a language is one instance of a much older problem — how anything gets into a human head and stays there — and that problem has been studied experimentally for a hundred and forty years. Some of what came out is unusually clear, unusually well replicated, and almost nobody uses it.
Each page below sets out one finding: what it claims, how strong the evidence is, what to do differently tomorrow, where it breaks down, and what it means for a language specifically. Where the field disagrees with itself, the page says so rather than picking the tidier answer.
Start with the biggest one → The reading list — 35 books, papers and courses (17 free)
Ways of practising that work
What to actually do with your study hours. These are the findings with the most evidence behind them, and the ones people are least likely to be using.
Deliberate practice
Mixed evidencepurposeful practice · the 10,000-hour rule (incorrectly)
Practice only improves you when it targets something you cannot yet do and tells you immediately whether you did it.
Elaboration
Good evidenceself-explanation · elaborative interrogation · the Feynman technique
Explaining why something is true, in your own words, turns a fact you have read into one you can use.
Interleaving
Good evidencemixed practice · varied practice
Mixing different kinds of problem in one session beats doing all of one kind, then all of the next.
Retrieval practice
Strong evidencethe testing effect · active recall · test-enhanced learning
Pulling something out of your memory strengthens it far more than putting it in again.
Spacing
Strong evidencedistributed practice · the spacing effect · spaced repetition
The same hours of study, spread over more days, produce far more durable memory than the same hours in one go.
Worked examples and cognitive load
Good evidencethe worked example effect · cognitive load theory · expertise reversal
Beginners learn more from studying full solutions than from struggling with problems — and experts learn less from them.
Why you misjudge your own learning
The reason knowing the techniques is not enough: the methods that work feel worse than the methods that do not.
Desirable difficulties
Good evidenceproductive struggle · storage strength vs retrieval strength
Some kinds of difficulty during study slow you down now and leave more behind later. Others are just difficulty.
Forgetting
Strong evidencethe forgetting curve · savings · retention interval
Forgetting is fast, mostly harmless, and the thing every good study schedule is built around.
Illusions of learning
Strong evidencefluency illusion · illusion of competence · judgments of learning
Your sense of how well you are learning tracks how easy the material feels right now, which is nearly the opposite of what it should track.
The conditions around the learning
Sleep and attention are not preliminaries to studying. They are part of how the studying works.
Attention and divided focus
Good evidencemultitasking · media multitasking · distraction
Nothing you did not attend to gets learned, and attention does not split — it alternates, expensively.
Sleep and memory
Good evidencememory consolidation · sleep-dependent consolidation
Sleep is not a pause between study sessions. It is when a large part of the learning actually happens.
Contested, and one that does not hold up
Widely repeated ideas whose evidence is weaker than their reputation — including the most popular belief in education, which fails the one test that would settle it.
Growth mindset
Contestedmindset theory · implicit theories of intelligence
Believing ability can grow does help — much less than the popular version claims, for some students, in some schools.
Learning styles
Does not hold upVAK · visual/auditory/kinaesthetic learners · the meshing hypothesis
People do have preferences about how material is presented. Matching teaching to those preferences does not improve learning.
Transfer and "learning to learn"
Does not hold upfar transfer · brain training · general thinking skills
Skills rarely carry across domains. What transfers is knowledge you already have, which is why there is no shortcut around learning things.
And for languages specifically
The findings here are about learning in general. The methods pages apply them — and disagree with them — in the particular case of a language: comprehensible input, spaced repetition, shadowing, structured courses, and the arguments between the people who advocate each.
And for a skill a machine can already do
The same findings, applied to the question that arrives now: how do you learn to design, to program, to write, to analyse, when a model produces a plausible version of the output in seconds? That is learning a skill in the age of AI — one page per skill, each with an explicit rule at every stage about what the model may and may not do, and what the research has actually measured.