This article by Dr Piotr Wozniak is part of SuperMemo Guru series on memory, learning, creativity, and problem solving.
Introduction
The way we formulate knowledge in learning has a monumental impact on memory. Thoughtfully crafted questions simplify learning, while well-organized knowledge ensures it stays with you for a lifetime.
Effective learning: 20 rules of knowledge formulation, compiled in 1999, describes the most important rules needed to effectively formulate good questions for long-term retention. Below, I provide a concise summary (esp. for those interested in incremental reading).
Summary: 20 rules
- don't learn things that you do not understand
- learn before you memorize
- start from basics before going into complexities
- keep questions simple
- pictures assist memory (see: video)
- learn mnemonic techniques, e.g. peg lists
- avoid lists, sets and enumerations (they can be tackled with cloze deletion as in the video)
- personalize and provide examples
- cloze deletion is fast and has a great mnemonic power
- use redundancy (similar items asking similar questions from different angles)(see: Knowledge Darwinism)
- use references
New rules: incremental reading
With the advent of incremental reading, the rules have been modified, and re-prioritized.
Changes of formulation strategy in incremental reading:
- building comprehension may be part of the learning process, and creating cloze deletions on poorly understood phrases is acceptable
- learning and memorization may occur in parallel (see: video)
- early in the learning process, items may be complex. They get simplified incrementally depending on knowledge priority, and the available time
- multiple cloze deletions on different formulations of the same statement may often substitute for mnemonic techniques (see: Knowledge darwinism)
- lists, sets, and enumerations can be easily tackled with cloze deletion (see: video)
- for personalization, add your own stories to texts that you learn. Elaborate creatively
A simple and universal litmus test for a good formulation is pleasure of learning. Each time you see a drop in pleasure, come back to this text and see if you can find a rule violation that might be responsible for the decline in fun.
We should strive at maximum applicability of knowledge.
Human intelligence is based on knowledge, of which abstract knowledge plays a particularly important role. Abstract knowledge is based on rules, such as 2+2=4. Such rules may be employed in multiple contexts and contribute to problem solving capacity. Rules are more useful than facts. For example, it is more useful to know that 2+2=4 (rule) than to know that a friends's phone number beings with 4 (fact). Rules and formulas are more applicable than facts. For more details see: Abstract knowledge
It is hard to keep high‑quality knowledge in mind if it is not used. The applicability rule, in part, implies you will use your knowledge in real life. Many problems tackled by the 20 rules go away once real‑life use comes into play. It provides the semantic glue for individual pieces. It resolves the problem of individual pieces siloed as atomic memories. The best users of spaced repetition associate each review with creative linking of previously learned pieces with actual or imaginary application in specific cases. This is why the problem mentioned in the Danger of flashcards is mitigated in users who are passionate about learning. They keep thinking about what they are learning. This is also why making repetitions in the morning, at one's creative peak, works best. Those who review in a tired state before sleep rarely wander with their thoughts beyond what a flashcard requires.
In representing knowledge, we should always strive at formulating atomic memories set in a good context of comprehension. If learning is enjoyable, items are probably formulated pretty well. The picture explains why simple memories are easier to retain:
Figure: Memory complexity illustrates the importance of the minimum information principle. When memorizing simple questions and answers, we can rely on a simple memory connection, and uniformly refresh that connection at review. Complex memories may have their concepts activated in an incomplete fashion, or in a different sequence that depends on the context. As a result, it is hard to produce a uniform increase in memory stability at review. Complex items are difficult to remember. An example of a simple item may be a word pair, e.g. apple = pomo (Esperanto). While a complex net of connection may be needed to recognize an apple. The connection between apple and pomo is irreducible (i.e. maximally simplified)
Parallels with AI
In experiments with AlterBrain we can observe a very similar preference for high-quality knowledge in an "artificial brain". Similar rules apply to build knowledge of high applicability. This allows of inference, which is a central component of intelligence.
A common misconception about the brain is that intelligence depends on the size of the brain or the speed of neurons. The entire IQ industry is based on that myth. Experiments with AlterBrain explain why quality knowledge is the decisive factor. The size of the brain is vastly beyond an intelligent man's needs. The speed of the brain varies to an uninteresting degree. In real life, even seconds do not matter if the creative outcome is of Einsteinian quality.
In humans and in machines, it is the curriculum that matters more for intelligence than capacity
LLMs need vast capacity for processing all human knowledge mixed up with a great deal of garbage. However, the ultimate model may be distilled and run on an average PC. Its usefulness will depend on the match between knowledge mastered and the demands of the task at hand.
Danger of flashcards
Experiments with AlterBrain demonstrate how perfect adherence to the 20 rules may still not be enough if you do not glue facts together. Your brain needs to build a concept network rich in semantics, generalization, and applicability. This is why the way you select your material (e.g. for high applicability) and the way you process it (e.g. with rich accompanying reasoning threads) may also determine how your material works out for you in real life. If you base AlterBrain on a small LLM and train it on a set of perfect flashcards, AlterBrain will start speaking "flashcardese". If you ask "Who is Vladimir Putin?", it might answer "a Russian politician". This is a highly semantically impoverished answer.
In addition, AlterBrain may lose on intelligence and increase hallucinations. After many years of schooling, due to the impact of 100 bad habits learned at school, a student's brain may also behave like a small LLM rich in overfitted facts.
This is why incremental reading is vastly superior to pure spaced repetition based on flashcards. Incremental reading demands learning in context. As long as learning is pleasurable, you can be reassured you are on the good way from cramming to actual high-quality learning and higher intelligence.
Sticking to the 20 rules is not enough. Learning needs to be semantic, intelligent, and pleasurable
For more texts on memory, learning, sleep, creativity, and problem solving, see Super Memory Guru