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

by Scott H. Young

During the MIT Challenge, I recognized that the most important resource for being able to eventually pass the classes wasn’t having access to recorded lectures, it was having access to problem sets.

directness isn’t an all-or-nothing feature but something you can gradually increase to improve your performance.

Transfer has been called the “Holy Grail of education.” It happens when you learn something in one context, say in a classroom, and are able to use it in another context, say in real life.

Given the problem of transfer and the importance of learning directly, let’s look at some of the ways that this is managed in different ultralearning projects. The simplest way to be direct is to learn by doing. Whenever possible, if you can spend a good portion of your learning time just doing the thing you want to get better at, the problem of directness will likely go away.

In Maini’s case, he wanted to be able to think and talk intelligently about machine learning, enough to be able to land a nontechnical role in a company that utilized those methods. That meant that being able to communicate his ideas articulately, understanding the concepts clearly, and being able to discuss them with both knowledgeable practitioners and laypeople was important. That’s why his goal to make a minicourse explaining the basics of machine learning fit so well. His learning was directly connected with where he wanted to apply the skill: communicating it to others.

Whereas the structures of our knowledge start out brittle, welded to the environments and contexts we learn them in, with more work and time they can become flexible and can be applied more broadly.

Many ultralearners opt for projects rather than classes to learn the skills they need. The rationale is simple: if you organize your learning around producing something, you’re guaranteed to at least learn how to produce that thing. If you take classes, you may spend a lot of time taking notes and reading but not achieve your goal.

Immersion is the process of surrounding yourself with the target environment in which the skill is practiced. This has the advantage of requiring much larger amounts of practice than would be typical, as well as exposing you to a fuller range of situations in which the skill applies.

Joining communities of people who are actively engaged in learning can have a similar impact, since it encourages constant exposure to new ideas and challenges.

It’s important to note that what matters for transfer is not every possible feature of the learning environment, such as what room you’re in or what clothes you’re wearing while you learn. Rather, it’s the cognitive features—situations where you need to make decisions about what to do and cue knowledge you’ve stored in your head. This suggests that when direct practice is impossible, a simulation of the environment will work to the degree to which it remains faithful to the cognitive elements of the task in question.

if you’re trying to evaluate what’s the best way to learn French before your trip to France, you’ll get more (although not perfect) transfer from doing Skype tutoring than you will from flipping through flash cards.

The last method I’ve found for enhancing directness is to increase the challenge, so that the skill level required is wholly contained within the goal that is set. Tristan de Montebello, when preparing to compete in the World Championship of Public Speaking, pushed to speak at middle schools, giving early versions of his talk. His feeling was that the feedback he received at Toastmasters clubs might be too soft or congratulatory to cut deep at what worked and didn’t work in his speech. Middle school students, in contrast, would be merciless. If a joke he said wasn’t funny or his delivery was boring or cheesy, he would be able to tell immediately from their faces what needed to be reworked. The overkill approach is to put yourself into an environment where the demands are going to be extremely high, so you’re unlikely to miss any important lessons or feedback.

You may feel as though you’re “not ready” to start speaking a language you’ve barely learned. You may be afraid to stand onstage and deliver a speech you haven’t memorized perfectly. You might not want to dive right into programming your own application and prefer to stick to watching videos where someone else does the coding. But these fears are often only temporary. If you can get enough motivation to start this method, it’s often a lot easier to continue it long term. The first week in each new country in my language learning project was always a shock, but soon it became completely normal to live entirely within the new language.

One way you can overkill a project is to aim for a particular test, performance, or challenge that will be above the ...

Another friend of mine decided to exhibit her photography as a means of pushing her skills and talent. Deciding in advance that your work will be viewable publicly alters your approach to learning and will gear you toward performance in the desired domain, rather than just checking off boxes of facts learned.

Learning, I’d like to argue, often works similarly, with certain aspects of the learning problem forming a bottleneck that controls the speed at which you can become more proficient overall.

Another rate-determining step could be vocabulary when learning a foreign language. The number of sentences you can successfully utter depends on how many words you know. If you know too few, you won’t be able to talk about very much. If you were able to suddenly inject hundreds of new words into your mental database, you might drastically expand your fluency even if your pronunciation, grammar, or other linguistic knowledge remains unchanged.

This is the strategy behind doing drills. By identifying a rate-determining step in your learning reaction, you can isolate it and work on it specifically. Since it governs the overall competence you have with that skill, by improving at it you will improve faster than if you try to practice every aspect of the skill at once. That was Franklin’s insight that allowed him to rapidly improve his writing: by identifying components of the overall skill of writing, figuring out which mattered in his situation, and then coming up with clever ways to emphasize them in his practice, he could get better more quickly than if he had just spent a lot of time writing.

The reason is that when you are practicing a complex skill, your cognitive resources (attention, memory, effort, etc.) must be spread over many different aspects of the task. When Franklin was writing, he had to consider not only the logical content of the argument he was making but word choice and rhetorical style. This can create a learning trap. In order to improve your performance in one aspect, you may need to devote so much attention to that one aspect that the other parts of your performance start to go down. If you can judge yourself only on how much you improve at the overall task, it can lead to a situation in which your improvement slows down because you will be getting worse at the overall task while becoming better at a specific component of it.

Drills resolve this problem by simplifying a skill enough that you can focus your cognitive resources on a single aspect.

When Franklin focused on reconstructing the order of an essay he had read previously, he could devote all his attention to asking what sequence of ideas leads to a good essay rather than also needing to worry abo...

A drill takes the direct practice and cuts it apart, so that you are practicing only an isolated component.

The tension between learning directly and doing drills can be resolved when we see them as being alternating stages in a larger cycle of learning.

Ultralearners, in contrast, frequently employ what I’ll call the Direct-Then-Drill Approach.

The first step is to try to practice the skill directly. This means figuring out where and how the skill will be used and then trying to match that situation as close as is feasible when practicing. Practice a language by actually speaking it. Learn programming by writing software. Improve your writing skills by penning essays. This initial connection and subsequent feedback loop ensure that the transfer problem won’t occur. The next step is to analyze the direct skill and try to isolate components that are either rate-determining steps in your performance or subskills you find difficult to improve because there are too many other things going on for you to focus on them. From here you can develop drills and practice those components separately until you get better at them. The final step is to go back to direct practice and integrate what you’ve learned. This has two purposes. The first is that even in well-designed drills, there are going to be transfer hiccups owing to the fact that what was previously an isolated skill must be moved to a new and more complex context. Think of this as being like building the connective tissue to join the muscles you strengthened separately. The second function of this step is as a check on whether your drill was well designed and appropriate. Many attempts to isolate a drill may end in failure because the drill doesn’t really cut at the heart of what was difficult in real practice. That’s okay; this feedback is important to help you min...

When writing a new article, you may have to juggle research, storytelling, vocabulary, and many other aspects simultaneously, making it hard to get a lot better at just one. Determining what to drill may seem tricky, but it doesn’t have to be. The key is to experiment. Make a hypothesis about what is holding you back, attack it with some drills, using the Direct-Then-Drill Approach, and you can quickly get feedback about whether you’re right.

The second difficulty with this principle is designing the drill to produce improvement. This is often hard because even if you recognize an aspect of your performance you’re weak on, it may be tricky to design a drill that trains that component without artificially removing what makes it difficult in actual application.

Teasing out the worst thing about your performance and practicing that in isolation takes guts.

Look for parts of the skill you’re learning that can be decomposed into specific moments of time that have heightened difficulty or importance.

Sometimes what you’ll want to practice isn’t a slice in time of a larger skill but a particular cognitive component.

When learning Mandarin Chinese, I would do tone drills that involved pronouncing pairs of words with different tones and recording myself speaking. That allowed me to practice producing different tones quickly, without the distraction of needing to remember what the words meant or how to form grammatically correct sentences.

copying the parts of the skill you don’t want to drill (either from someone else or your past work), you can focus exclusively on the component you want to practice. Not only does this save a lot of time, because you need to repeat only the part you’re drilling, it also reduces your cognitive burden, meaning you can apply more focus to getting better at that one aspect.

Drilling problems without context is mind-numbing. However, once you’ve identified that it’s the bottleneck preventing you from going further, they become instilled with new purpose. In ultralearning, which is directed by the student, not an external source, drills take on a new light. Instead of being forced to do them for unknown purposes, it is now up to you to find a way to enhance the learning process by accelerating learning on the specific things that you find most difficult. In this sense, drills take on a very different flavor in ultralearning as opposed to traditional learning.

The difficulty and usefulness of drills repeat a pattern that will recur throughout the ultralearning principles: that something mentally strenuous provides a greater benefit to learning than something easy. Nowhere is this pattern more clear than in the next principle, retrieval, where difficulty itself may be the key to more effective learning.

It pays better to wait and recollect by an effort from within, than to look at the book again. —William James, psychologist

Imagine you’re a student preparing for an exam. You have three choices about how you can allocate your limited studying time. First, you can review the material. You can look over your notes and book and study everything until you’re sure you’ll remember it. Second, you can test yourself. You can keep the book shut and try to remember what was in it. Finally, you can create a concept map. You can write out the main concepts in a diagram, showing how they’re organized and related to other items you need to study. If you can pick only one, which one should you choose to do best on the final exam?

Testing yourself—trying to retrieve information without looking at the text—clearly outperformed all other conditions.

In this experiment the final test was to produce a concept map. Despite the overwhelming similarity to the evaluation task, free recall still did better than using concept mapping to study.

Another possible explanation for why self-testing works is feedback. When you review something passively, you don’t get any feedback about what you know and don’t know. Since tests usually come with feedback, that might explain why students who practiced self-testing beat the concept mappers or passive reviewers. Though it is true that feedback is valuable, once again, retrieval doesn’t simply reduce down to getting more feedback. In the experiments mentioned, students were asked to do free recall but weren’t provided any feedback about items they missed or got wrong. The act of trying to summon up knowledge from memory is a powerful learning tool on its own, beyond its connection to direct practice or feedback.

Without the answers at hand, Ramanujan was forced to invent his own solutions to the problems, retrieving information from his mind rather than reviewing it in a book.

Karpicke’s research points to a possible explanation: Human beings don’t have the ability to know with certainty how well they’ve learned something. Instead, we need to rely on clues from our experience of studying to give us a feeling about how well we’re doing. These so-called judgments of learning (JOLs) are based, in part, on how fluently we can process something. If the learning task feels easy and smooth, we are more likely to believe we’ve learned it. If the task feels like a struggle, we’ll feel we haven’t learned it yet. Immediately after spending some time studying, these JOLs may even be accurate. Minutes after studying something using a strategy of passive review, students perform better than they would if they had practiced retrieval. The feeling that you’re learning more when you’re reading rather than trying to recall with a closed book isn’t inaccurate. The problem comes after. Test again days later, and retrieval practice beats passive review by a mile. What helped in the immediate time after studying turns out not to create the long-term memory needed for actual learning to take place.

Another explanation for why students opt for low-efficiency review instead of retrieval is that they don’t feel they know the material well enough to test themselves on it. In another experiment, Karpicke had students choose a strategy for learning. Inevitably, students who were performing more weakly elected to review the material first, waiting until they were “ready” to start practice testing. If through experimental intervention, however, they were forced to practice retrieval earlier, they learned more. Whether you are ready or not, retrieval practice works better. Especially if you combine retrieval with the ability to look up the answers, retrieval practice is a much better form of studying than the ones most students apply.

What makes practicing retrieval so much better than review? One answer comes from the psychologist R. A. Bjork’s concept of desirable difficulty. More difficult retrieval leads to better learning, provided the act of retrieval is itself successful.

Difficulty, far from being an obstacle to making retrieval work, may be part of the reason it does so.

Low-intensity learning strategies typically involve either less or easier retrieval. Pushing difficulty higher and opting for testing oneself well before you are “ready” is more efficient.

One can think back to Benny Lewis’s strategy of speaking a new language from the first day. Though this approach is high in difficulty, research suggests why it might be more useful than easier forms of classroom study. Placing himself in a more difficult context means that every time Lewis needs to recall a word or phrase, it will be remembered more strongly than when doing the same act of retrieval in a classroom setting and much better than when simply looking over a list of words and phrases.

Difficulty can become undesirable if it gets so hard that retrieval becomes impossible. Delaying the first test of a newly learned fact has some benefits over testing immediately. However, if you delay the test too long, the information may be forgotten entirely. The idea, therefore, is to find the right midpoint: far enough away to make w...

the act of taking a test not only is a source of learning but results in more learning than a similar amount of time spent in review. However, this still fits within the conventional idea of knowledge being first acquired, and then strengthened or tested later.

An interesting observation from retrieval research, known as the forward-testing effect, shows that retrieval not only helps enhance what you’ve learned previously but can even help prepare you to learn better. Regular testing of previously studied information can make it easier to learn new information. This means that retrieval works to enhance future learning, even when there is nothing to retrieve yet!

Some researchers argue that it may be that trying to find knowledge that hasn’t been learned yet—say, by trying to solve a problem you haven’t learned the answer to yet—nonetheless helps reinforce search strategies that are put to use once the knowledge is encountered later. An analogy here is that trying to retrieve an answer that doesn’t yet exist in your mind is like laying down a road leading to a building that hasn’t been constructed yet. The destination doesn’t exist, but the path to get to where it will be, once constructed, is developed regardless.