Ultralearning

His definition: a strategy for acquiring skills and knowledge that is self-directed and intense. Self-directed means you make the decisions about what and how; intense means the approach prioritises effectiveness over comfort.

His own projects — completing a computer science curriculum's material in a year, learning several languages in quick succession — are the evidence, and he is reasonably clear that they are demonstrations rather than proof.

Self-directed and intense. He is clear his own projects are demonstrations rather than evidence.

Devmark’s reading — not a quotation from the book

His first principle and the one he considers most important. Before learning, research how the skill is structured, what the components are, how others have learned it, and what the standard failure modes are.

His rule of thumb is to spend roughly ten per cent of the total project time on this, on the reasoning that learning the wrong material efficiently is the most expensive error available.

His most useful single principle: learn by doing the thing you actually want to be able to do, in the context you want to do it.

His argument is that transfer between contexts is far weaker than people assume — studying a language from a textbook transfers poorly to conversation, and studying management theory transfers poorly to managing. The remedy is to make the practice resemble the target situation.

His treatment of isolating components: identify the rate-limiting part of the skill and practise it separately at higher intensity than the whole activity would allow.

He pairs this with a warning — drilling is only useful if the component is then reintegrated, because isolated competence that never returns to the real task is the classic failure of academic practice.

His most research-supported principle. Testing yourself produces substantially better retention than re-reading, even when the testing feels less productive and even before you have learned the material properly.

The counter-intuitive finding he emphasises is that attempting to retrieve something you cannot yet recall improves subsequent learning of it, which means testing can precede study rather than following it.

Trying and failing to recall something improves how well you later learn it. Testing can come before studying.

Devmark’s reading — not a quotation from the book

He separates outcome feedback (you did badly), informational feedback (here is what went wrong), and corrective feedback (here is what to do instead) — arguing the third is rare, most valuable, and usually requires a person.

His practical advice is to aggressively seek the uncomfortable kind, and his observation that people optimise for feedback that is pleasant rather than informative is well made.

His later principles cover spacing and the forgetting curve, building deep intuition by insisting on explaining things in plain terms rather than memorising, and eventually moving from following methods to designing your own.

The Feynman technique — explain it simply and notice exactly where you stall — is the most immediately usable tool in this section.

The evidence is a mixture of solid learning science — retrieval practice, spacing, interleaving are well established — and the author's own project reports, which are compelling and uncontrolled.

The method also demands time availability and a tolerance for discomfort that most people cannot sustain alongside employment, which the book acknowledges only briefly.

Substantial overlap with *Make It Stick* on the cognitive science and with Ericsson on deliberate practice. Young's contribution is packaging these into a project methodology rather than discovering anything.

If you want the research, read *Make It Stick*. If you want a procedure for running a self-directed learning project, this is the better book.

Directness is the highest-value principle and the most commonly violated: check whether your practice resembles the situation you are preparing for, and if it does not, change the practice rather than doing more of it.

And retrieval over review, which is the best-supported finding in learning science and the one almost nobody applies because re-reading feels productive and testing does not.