Ultralearning: Master Hard Skills, Outsmart the Competition, and Accelerate Your Career

The summary

In October 2011 Scott Young opened MIT’s OpenCourseWare and started working through the computer science degree. Twelve months later he had finished 33 classes — the same credit load as the four-year program — writing the final exam for each one and grading it against MIT’s own answer key. He never enrolled, never earned the degree, never had a professor. Ultralearning is what came after: he went looking for other people who learn like this, found them scattered across languages, game shows and public speaking, and pulled out the nine things they do in common. The strategy is self-directed, because you build the curriculum, and aggressive, because at every fork you take the version of practice that’s harder and less pleasant.

What you learn in a classroom mostly stays there

Young’s starting complaint is transfer. The research keeps finding that much of what we set out to learn never reaches the situations we needed it for — students earn high marks in a subject and then can’t apply it once the context shifts even slightly. Years of French, no conversation. A semester of statistics, no ability to read a study.

His antidote is directness: practice the thing you actually want to be able to do. A learning project, he argues, should involve something you do or make, not just something you study. Want French? Have conversations in French. Want to build websites? Build one now, badly.

He tested this on himself in September 2013, when he and his friend Vat Jaiswal spent a year abroad under a single rule: no English. Roughly three months each in Spain, Brazil, China and Taiwan, and South Korea, with English permitted only for under an hour of work a week, calls home, and emergencies. They came out at an intermediate level in all four languages, having broken the rule occasionally in China and Korea, where the difficulty and the fatigue caught up with them. The itinerary isn’t the lesson. Immersion is just directness with nowhere to hide.

Map the territory, then attack the bottleneck

Before Roger Craig went on Jeopardy!, he treated the show as a data problem. He pulled down the J! Archive of past clues, used data mining and text clustering to work out which topics actually come up, built himself a tool to quiz on them, and watched where his answers fell apart. On September 14, 2010, he won $77,000 in a single game, taking Ken Jennings’s record by two thousand dollars. His seven-game run brought $231,200, and he won the Tournament of Champions the year after.

That front-loaded research is what Young calls metalearning: work out how the skill is structured, what has to be mastered, and how other people got there, before you spend a single hour grinding. It’s the step that decides whether everything else is aimed at anything. Focus, the principle beside it, is far less glamorous — hard skills need concentrated hours, and Young has since said the distraction problem has only worsened since he wrote the book.

Once the map exists, drill. Break the skill into components, find the one holding the whole thing back, work it in isolation, then fold it straight back into real practice so you can check whether the gain survives contact with the task. Nigel Richards is the extreme case: in 2015 he won the French-language World Scrabble Championship 565–434, after eight or nine weeks spent memorizing the French dictionary from word lists and dictionary pages. He doesn’t speak French. He drilled one component of the game to the exclusion of the language it’s played in.

A test is a tool, not a verdict

Most people treat a quiz as a measurement — the thing that tells you whether the studying worked. Ultralearners treat it as the studying. Craig’s spaced-repetition decks in Anki weren’t checking his progress; dragging an answer out of memory is how the answer got fixed there in the first place. Rereading your notes feels productive and does little. Recalling them with the notes shut feels like failing and does most of the work.

Feedback gets the same treatment, and Young’s rule of thumb is that it’s useful roughly in proportion to how much it stings — a heuristic the research doesn’t really bear out, though the underlying advice to go and get some is sound. Tristan de Montebello had close to no speaking experience when he started. Seven months and about 500 hours later he was a finalist at the World Championship of Public Speaking, on stage in front of 3,000 people. He got there by giving 80 to 100 speeches to live audiences — Toastmasters, schools, improv groups, strangers — and running material past his wife, movie directors and improv teachers long before it reached a stage. Comedians work this way because they have to. Most learners never build the loop at all.

Nobody hands you the curriculum

The last principles — retention, intuition, experimentation — all point the same way: sooner or later the guidance runs out and you have to run your own trials. Eric Barone spent roughly four and a half years alone on Stardew Valley, doing the design, programming, art, animation, music and writing himself. He came in with a computer science degree and taught himself the rest as he went — sinking hundreds of hours into pixel art he found genuinely hard, and scoring the game with no formal musical training. He redid the artwork over and over because nobody else was going to tell him when it was finished.

The bottom line

Learning something hard, fast, is a set of choices rather than a gift, and nearly all of those choices cut against instinct: do the real thing before you feel ready, test yourself instead of reviewing, go looking for the feedback you’d rather not hear. The projects in this book look extreme because the method is uncomfortable, not because the people running them were unusually gifted. Read it if you’ve committed to learning something difficult on your own and want a plan sharper than “sign up for a course.”