Learning How to Learn

Meta as hell, but an important concept that everybody should learn to use on a daily basis.

I. Meta-Learning

Meta-learning is the deliberate development of one’s own capacity to learn. The work is one level up from any specific learning task; it concerns how you approach learning generally rather than what you happen to be learning at the moment.

Hacks are tactical refinements; meta-learning is the broader practice that determines whether the hacks produce sustained capability development or just produce the feeling of being productive.

Most adults have unmeta-learned learning. They acquired their learning approach during schooling, never examined it, and continue to apply it unchanged decades later. The approach may have worked tolerably well for some school contexts; it often doesn’t work for adult learning needs. The meta-learning work involves noticing what your current approach is, evaluating what it produces, and modifying it deliberately.

This is the skill the schooling system specifically failed to teach. Schooling delivers content; meta-learning is the practice that helps you acquire content efficiently. The mismatch is part of why so many adults find learning harder than it needs to be. The capacity is genuinely there; the practice for accessing it has not been developed.

II. The Dreyfus Skill Acquisition Model

The foundational framework for understanding what skill development actually looks like. Developed by Hubert and Stuart Dreyfus in the 1980s based on observations of pilots, chess players, and other domain experts. Originally adopted in nursing education in the early 1980s; subsequently applied across multiple domains.

The five stages describe how capability develops in any specific domain, with implications for what each stage requires.

• Stage 1: Novice: Little to no experience. Operates from explicit rules and instructions. Cannot judge what the rules don’t cover. Needs recipes rather than understanding. The call centre worker reading a script is operating at the novice level by design; the work is structured to require minimal underlying competence.
  • Novices are concerned with their ability to succeed. With little experience to guide them, they don’t know whether their actions will turn out correctly. They are vulnerable to confusion when situations don’t fit the rules. They typically don’t want to learn; they want to accomplish the task. Learning is painful at this stage.
  • The problem with rules is that they can be misinterpreted and vague. This is what’s sometimes called infinite regression. Rules get you started, but cannot take you further.
• Stage 2: Advanced Beginner: Can break away from the fixed rules. Still has problems troubleshooting. Wants information fast and doesn’t want to be bogged down with detailed explanations. Doesn’t want the big picture and may dismiss it as irrelevant if forced. No holistic understanding yet, and not seeking one.
  • Advanced beginners can perform many tasks adequately but struggle with novel situations. They have started building pattern recognition, but cannot yet articulate the patterns.
• Stage 3: Competent: Practitioners can develop conceptual models of the problem domain and work with these models effectively. Can begin to seek out and apply advice from experts. No longer knee-jerk responses; will solve problems based on previous experience. However, still difficulty without fine details to focus on.
  • These people are usually the ones who “have intuition” and are “being resourceful.” Typically found in leadership positions where they can teach novices and try to avoid annoying the experts.
• Stage 4: Proficient: Needs the big picture. Wants to learn the framework. Frustrated by oversimplified information. Can correct poor task performance. Reflects on past performance and revises the approach to perform better next time. Can read case studies and learn from others. Learns to test everything that might break.
  • Proficient practitioners can teach more because they have the conceptual structure to organise their teaching. They have moved from “doing the work” to “understanding the work.”
• Stage 5: Expert: The primary sources of knowledge and information in any field. Continually looking for better methods and ways of doing things. Write the books, the articles, and do the lecture circuits. Approximately 1-5% of the population in any specific domain.
  • Experts work from intuition rather than reason. A combination of subtle cues gives them the answer; they cannot always articulate why. They are very good at targeted, focused pattern matching.
  • When you make experts write rules for novices, it helps the novice. When you make experts follow their own rules, they lose their expert edge. You cannot herd a racehorse. Rules for novices and intuition for experts; the two stages require different approaches.

III. The Second-Order Incompetence Problem

The Kruger-Dunning research: Justin Kruger and David Dunning’s 1999 paper “Unskilled and Unaware of It: How Difficulties in Recognizing One’s Own Incompetence Lead to Inflated Self-assessments” documented what has become widely known as the Dunning-Kruger effect.

The basic finding: people with low ability in a domain often overestimate their competence. They lack the metacognitive capacity to recognise their own incompetence because the same skills required to be competent are required to evaluate competence. The Kruger-Dunning paper opened with a story about a thief who robbed a bank in broad daylight, wearing lemon juice on his face, confident that the juice would make him invisible. This is second-order incompetence: not knowing what you don’t know.

Charles Darwin’s observation captures the same pattern: “Ignorance more frequently begets confidence than does knowledge.”

Once you become an expert, you become painfully aware of just how little you really know. The expert’s epistemic humility is partly an occupational adaptation; you have done enough work to discover the boundaries of your knowledge.

The implications for meta-learning:

• Most lower-stage learners overestimate their abilities by as much as 50%
• Metacognitive abilities (self-awareness of one’s own thinking) are only fully accessible at higher skill levels
• The novice cannot reliably evaluate their own progress; they need external feedback to advance
• The early stages of any new domain require humility that doesn’t come naturally because you don’t yet know what you don’t know
• Looking up the actual research on Kruger-Dunning produces a more accurate picture than the popular oversimplifications; the effect has been substantively contested in some specific framings, but the general pattern holds

When entering a new domain, assume you are less competent than you feel. Seek out genuine experts and engage with their feedback even when (especially when) the feedback is uncomfortable. The instinct to defend your initial competence assessment is exactly the second-order incompetence operating; recognising it is part of moving past it.

IV. The Linear vs Rich Mode Framing

A useful framework drawn from Andy Hunt’s Pragmatic Thinking and Learning.

Linear mode (L-mode): The deliberate, analytical, verbal processing that handles step-by-step work. Slow. Uses limited cognitive resources. Operates through words, symbols, and explicit logic. Equivalent in some ways to Kahneman’s System 2 but framed around process rather than speed.

L-mode capacities:

• Verbal: using words to name, describe, and define
• Analytic: figuring things out step-by-step and part-by-part
• Symbolic: using symbols to stand for things
• Abstract: taking out small bits of information to represent the whole
• Temporal: keeping track of time and sequencing
• Rational: drawing conclusions based on reason and facts
• Digital: using numbers, as in counting
• Logical: drawing conclusions based on theorems and well-stated arguments
• Linear: thinking in terms of linked ideas, one thought directly following another

Rich mode (R-mode): The holistic, intuitive, pattern-matching processing that handles non-verbal cognition. Operates spatially and in relation to other things. Makes leaps from insight. Sometimes operates non-rationally based on hunches.

R-mode capacities:

• Non-verbal
• Synthetic (combining elements rather than separating them)
• Concrete
• Analogic (recognising similarities and differences through analogy)
• Non-rational (not anti-rational; not bound by explicit rules)
• Spatial
• Intuitive
• Holistic (seeing the whole rather than the parts)
• Non-linear

When L-mode is being used, R-mode cannot get access. They interfere with each other. You need both. R for intuition, problem-solving, and creativity; L for working through details and making things happen.

Western culture has tended to favour L-mode and treat R-mode as the lesser of mortals. The bias has been adopted in education systems that emphasise testable analytical performance over creative integration. The bias has costs: overreliance on L-mode loses access to R-mode capabilities.

The L-mode/R-mode distinction oversimplifies the underlying neuroscience. Both hemispheres engage in both analytical and holistic processing in complex distributed networks. Treat the L/R framing as a useful metaphor rather than as accurate neuroanatomy.

Different learning tasks benefit from different modes. Mathematical problem-solving leans L-mode; creative integration leans R-mode; learning typically requires both at different stages. Noticing which mode is appropriate for the current work and deliberately engaging it produces better outcomes than defaulting to whichever is your habitual mode.

V. The Optimal Stress State for Learning

• Too little stress: No challenge, no engagement. The system doesn’t recruit the neurochemistry that supports learning. Boredom prevents engagement.
• Too much stress: The threat response activates. Cortisol elevation impairs hippocampal function. The body shifts into survival mode. Learning capacity collapses. The chronic stress of poorly-designed schooling environments produces predictable learning impairment.
• Optimal stress: Sufficient challenge to engage attention and produce the neurochemistry of learning. Insufficient threat to trigger survival response. The “challenge-skill match” that Csikszentmihalyi described as flow.

The neurochemical substrate:

• Adequate alertness from norepinephrine and epinephrine
• Focused attention from acetylcholine release
• Reward signalling from dopamine when errors are corrected
• The willingness to make errors, which the dopamine reward depends on
• Subsequent consolidation during sleep

The combination requires being calm enough to focus but activated enough to engage. The state is sometimes called “alert calm” or “focused engagement.”

The interventions for accessing the optimal state:

• Adequate sleep (see Sleep & Circadian Rhythm)
• Adequate nutrition (see Nutrition)
• Breath-based state regulation (see Breathing)
• Caffeine timed appropriately (see Habit Building Cheatsheet)
• Environment design for focus (see Environment)
• Movement before sustained learning (see Movement)

Trying to learn well while sleep-deprived, malnourished, dysregulated, or in a hostile environment produces predictably poor outcomes regardless of the learning techniques applied.

VI. The 85% Rule Applied to Learning

The finding from learning research is that 85% success rate is approximately optimal for sustained engagement and development.

• The mechanism: Errors are the entry point for neuroplasticity. The state of frustration cues a greater number of brain areas to be more alert, so subsequent attempts at learning have higher focus and a greater probability of acquiring the new information or skill. Without errors, the system doesn’t engage the learning machinery.
• Why too easy fails: Making a goal too easy means no growth. The system isn’t being challenged enough to engage neuroplasticity. You succeed but don’t actually develop new capability. Self-esteem might feel pleasant, but you’re not building anything useful.
• Why too hard fails: Making a goal too lofty means no progress and eventual quitting. The system is overwhelmed; the neuroplasticity that would have engaged at moderate challenge doesn’t activate at extreme challenge. You fail repeatedly without learning from the failures.
• The 85% sweet spot: Challenge level where you’re succeeding most of the time but failing enough to engage neuroplasticity. The 15% failure rate provides the error-driven learning that produces development.

Applying it:

• For deliberate practice in any skill, calibrate the difficulty to 85% success
• When tasks become too easy, increase the difficulty until you return to 15% failure
• When tasks become too hard, scale back until 85% success returns
• Track the calibration deliberately rather than letting it drift

Schooling typically operates at less than 85% success expectation. Standard grading often punishes failures rather than treating them as learning opportunities. The 15% failure rate that would optimise learning is treated as below-acceptable performance. The mismatch is part of what produces sub-optimal learning outcomes in formal education.

VII. Recognising Diminishing Efficiency

A meta-learning skill: noticing when continued effort is producing diminishing returns.

The patterns:

• Re-reading the same material without comprehension increasing
• Attempting practice that no longer improves performance
• Spending more time on a task than the task warrants
• Fatigue accumulating without proportional output
• Attention drifting despite trying to focus
• Subsequent practice producing worse rather than better results

The mechanisms:

• Cognitive fatigue from sustained working memory use
• Saturation effects where additional input doesn’t integrate
• The need for sleep consolidation before further progress
• Insufficient recovery between learning bouts
• Misaligned tools or methods for current stage

The interventions:

• Stop the current session and rest
• Switch to different material that engages different cognitive resources
• Take a walk or do mild physical activity
• Do a NSDR (non-sleep deep rest) protocol
• Sleep, particularly if you’re approaching the end of the day
• Return to the material after consolidation rather than pushing through

Western productivity culture often celebrates pushing through diminishing returns. The “grind” framing treats sustained low-output effort as virtuous. The learning research suggests the opposite: stopping when efficiency drops produces substantially better long-term outcomes than continuing to grind. The grind produces the feeling of working hard; it doesn’t reliably produce learning.

Most people have an impaired ability to notice when their learning is no longer effective. The schooling experience trained sustained effort regardless of outcome; the trained pattern persists. Reclaiming the noticing requires explicit attention to internal state during learning rather than just to external task completion.

VIII. Motivation Architecture for Learning

Motivation is foundational to sustained learning.

Intrinsic vs extrinsic motivation: Intrinsic motivation (engaging because the activity itself is rewarding) produces more durable engagement than extrinsic motivation (engaging for external reward). For sustained learning across years, intrinsic motivation is foundational.

The autonomy, competence, relatedness framework: Edward Deci and Richard Ryan’s Self-Determination Theory identified three components that support intrinsic motivation:

• Autonomy: Sense of choice and self-direction in the activity
• Competence: Sense that you can succeed at the activity through your own effort
• Relatedness: Sense of connection with others through the activity

When all three are present, motivation typically sustains. When one or more are missing, motivation typically degrades. Schooling often undermines all three (assigned curriculum, frequent failure experiences, isolated assessment), which contributes to the learning aversion the system produces.

The implementation for adults:

• Choose what to learn rather than waiting to be told
• Calibrate difficulty to support competence (the 85% rule)
• Find others who share the learning interest

Curiosity is the felt sense of wanting to understand. The neurochemistry involves dopamine release in response to anticipated information gain. Sustained curiosity is one of the more reliable substrates for ongoing learning.

The curiosity orientation can be developed deliberately:

• Notice what you genuinely wonder about
• Allow yourself to investigate those wonderings without justifying the investigation
• Treat “I don’t know” as an invitation rather than as a failure
• Follow the questions across disciplines rather than staying within authorised territory

Some learners are motivated primarily by fear of consequences (failing, being judged, falling behind). The motivation works in the short term but produces predictable outcomes: shallow engagement, brittle retention, sustained anxiety around learning, and eventual avoidance. The fear-motivated learning is what schooling often produces; reclaiming discovery requires shifting toward curiosity-motivated learning.

IX. Collaboration in Learning

Learning happens better in collaboration than in isolation.

What collaboration provides:

• Multiple perspectives on the material
• Social accountability for engagement
• Opportunity to teach and be taught
• Feedback on understanding through verbal exchange
• Emotional regulation through shared experience
• Pattern recognition through diverse examples
• Motivation through peer engagement

What collaboration doesn’t always provide:

• Better learning than focused solo work in some specific contexts
• Reliable knowledge transfer if the collaborators are at similar levels of incompetence
• Substantive engagement if the social dynamic is performative
• Genuine progress if collaboration substitutes for the difficult solo work that some learning requires

The optimal pattern: Most learning involves a mix of solo work and collaborative work. The solo work allows sustained engagement with difficult material; the collaborative work provides perspective, feedback, and motivation. The proportion depends on the specific learning task and the individual’s preferences.

The peer-effect research: Multiple studies establish that peer composition affects learning outcomes. Being around higher-performing peers tends to lift performance; being around lower-performing peers tends to depress performance. The mechanism is partly modelling, partly expectations, partly access to better explanations.

Adult learning:

• Seek learning communities with engagement rather than performative engagement
• Look for peers at similar or slightly higher levels who are doing the work
• Be willing to ask basic questions in safe contexts, even when it surfaces incompetence
• Teach others as part of the learning rather than waiting until you’ve “finished” learning

The traditional reading group (book, regular meetings, sustained engagement over weeks or months) is one of the more reliable learning structures for adults outside formal education. The structure provides accountability, social engagement, perspective diversity, and depth through sustained focus.

X. Outsourcing Mental Power

The externalised-cognition work covered in Brain 2.0.

The biological brain has constraints (working memory limits, retrieval failures, forgetting curves). The constraints are not removable; they evolved features rather than bugs. The compensation move is externalising parts of cognition through writing, notes, databases, and broader tools.

Building Brain 2.0 infrastructure is part of meta-learning rather than separate from it. The infrastructure changes what you can think with; the change supports subsequent learning that would not have been possible without the infrastructure.

The integration:

• Capture is the meta-learning version of attention deployment
• Organisation is the meta-learning version of categorisation
• Connection is the meta-learning version of integration
• Retrieval is the meta-learning version of recall
• Active use is the meta-learning version of the application

The Brain 2.0 work and the meta-learning work are aspects of the same broader project: developing your capacity to learn across the years.

The AI extension question: Covered in Brain 2.0. The LLM tools change what externalisation can do; they also introduce new failure modes (sycophancy, hallucination, atrophy of independent reasoning). Useful tools that require careful integration rather than wholesale adoption. Don’t give in to the convenience, as there are no free lunches. 

XI. The Power of Summarising

Summarising material in your own words improves retention and understanding. The act of compression forces you to identify what matters and articulate it. The articulation tests your understanding; the test reveals gaps you can then address.

Passive reading produces limited learning. Active engagement with material (writing about it, summarising it, teaching it) produces more learning. Summarising is one of the more accessible active-engagement methods.

The SQ3R framework (from Andy Hunt): 

• Survey: Scan the table of contents and chapter summaries for an overview
• Question: Note any questions you have before reading deeply
• Read: Read the material in entirety
• Recite: Summarise the material, take notes, put it in your own words
• Review: Reread your notes, expand them, discuss with colleagues

The progressive summarisation alternative: Tiago Forte’s method of progressively narrowing through highlighting, bolding, italicising, then writing a summary at the top of the note. Different in form from SQ3R; similar in principle.

The practical implementation:

• When reading anything you actually want to learn, summarise it before moving on
• The summary should be in your own words, not copied phrases
• The summary should be shorter than the original; compression forces understanding
• Return to your summary later and check whether you can reconstruct the material
• If you can’t, the summary needs improvement, or the underlying engagement was insufficient

Most reading is consumption rather than learning. The summary discipline converts more of your reading into learning. The cost is sustained attention; the benefit is better retention and understanding.

XII. Learning by Teaching

The Feynman Technique covered in Mental Model Basics applied to learning generally.

If you cannot explain something simply, you don’t understand it. The act of teaching surfaces gaps in your understanding that passive consumption hides.

Teaching requires you to:

• Identify what the learner doesn’t know
• Sequence the material appropriately
• Find examples and analogies that work
• Anticipate questions and confusion points
• Test the learner’s understanding
• Revise your explanation when it fails

All of these require deeper engagement with the material than simply reading it. The depth produces durable learning that simpler consumption doesn’t.

The implementation forms:

• Actually teach the material to someone who doesn’t know it
• Write a tutorial as if for a beginner
• Explain the material to yourself out loud
• Imagine teaching a specific person you know who would have specific questions
• Record yourself explaining the material and listen back
• Tutor others in the material you’re learning

Two people learning the same material can take turns teaching specific portions to each other. The arrangement produces better learning than either person studying alone. The mutual teaching creates accountability and surfaces gaps in both partners’ understanding.

Teaching requires having something to teach. The Feynman Technique applied prematurely (before you have enough understanding to teach anything) produces theatre rather than learning. The reasonable sequence is to study first, then teach; the teaching reinforces and tests the study rather than substituting for it.

XIII. The Beginner’s Mind Question

A useful orientation drawn partly from contemplative traditions. Often associated with Zen Buddhism through Shunryu Suzuki’s Zen Mind, Beginner’s Mind (1970).

• The principle: In the beginner’s mind, there are many possibilities; in the expert’s mind, there are few. The beginner approaches material without preconceptions; the expert often approaches it with strong preconceptions that filter out useful information.
• The expertise trap: Once you have developed competence in a domain, you often stop being a learner. You “know how this works,” and the existing knowledge filters out information that would update the model. Lapsed experts become stale within their own field because they stop noticing what could disconfirm their existing framing.
• The practical implication: Maintain a beginner’s mind even within domains where you have expertise. Continue to ask basic questions. Continue to engage with new entrants in the field. Continue to test your existing framings against new evidence. The beginner’s posture is not the same as ignorance; it is the willingness to be wrong combined with existing knowledge.
• The cross-domain version: When entering a new domain, the beginner’s mind is required by circumstance. When the new domain shares features with one you already know, there’s a temptation to import the existing framing wholesale. Sometimes the import works; often it produces the second-order incompetence pattern (you don’t know what’s specifically different about the new domain). The reasonable approach treats each new domain as novel until evidence suggests otherwise.
• The Suzuki calibration: Zen traditions have produced value alongside specific community dynamics. Suzuki’s Zen Mind, Beginner’s Mind is one of the more accessible contemporary introductions; the broader Zen institutional context has its own complexities.

XIV. Common Failure Modes

• The collector’s pattern: Accumulating learning methods without applying them. Reading books about how to learn while not actually learning. The meta-level becomes the substitute for the underlying activity. Cure: pick a small set of methods that suit your style and commit to them for at least 90 days before evaluating.
• The optimisation trap: Spending more time on the meta-learning system than on the learning. Tools serve learning; learning does not serve tools. Cure: simplest system you’ll use beats elaborate system you’ll abandon.
• The grind pattern: Pushing through diminishing returns because the cultural framing celebrates sustained effort. Cure: notice the efficiency decline and respond to it rather than overriding the signal.
• The novelty addiction: Constantly switching to new methods, looking for the perfect approach. Cure: commit to a reasonable method long enough to evaluate it; switching too frequently prevents seeing what works.
• The expertise trap: Becoming so skilled in one domain that you stop being a learner. Cure: maintain beginner’s mind; engage with new entrants in your field; periodically test your existing framing against new evidence.
• The collaboration substitution: Using collaborative learning to avoid the solo work that learning sometimes requires. Cure: notice when collaboration is genuine engagement versus a social substitute for difficulty.
• The credential pursuit: Optimising for credentials rather than capability. Cure: distinguish between what’s required for legitimate gatekeeping versus what produces actual capability; weight capability development accordingly.
• The motivation-from-fear pattern: Learning primarily because of fear of consequences. Produces shallow engagement and eventual avoidance. Cure: deliberate work to shift toward curiosity-based motivation; this is part of what the broader Discovery section addresses.
• The information consumption pattern: Substituting information consumption for active learning. Cure: build expression into your learning practice; if you’re not producing anything from what you’re consuming, you’re not really learning.

XV. Cross-Links

• Discovery for the section overview
• Discovery Basics for the underlying neuroscience
• Rebranding Learning for the schooling critique
• Learning Cheatsheet for the practical catalogue
• The Learning Rabbit Hole for the deeper material
• Resources for the reading list