Cognition methods¶

Cognition methods are external scaffolds humans use to push a small, leaky, generative brain past its native limits. Most reduce to a handful of mechanisms — spaced retrieval, deliberate cueing, chunking, imagery, offloading, and dialogue. History recorded the same tricks across cultures (Simonides, Ricci, Luhmann, Polgar) because the underlying brain is the same. The frontier is multi-expert cooperation: running several methods, several voices, or several selves on the same problem concurrently, with explicit arbitration.

🌿 budding tended 2026-05-14 research cognition learning memory methods history multi-expert

flowchart LR
  brain[finite generator · 3-7 slots] --> meth[method scaffold]
  meth --> mem[stronger encoding]
  meth --> recall[reliable cue]
  meth --> mix[wider sampling]
  meth --> ext[external trace]
  mem & recall & mix & ext --> out[durable skill · creative output]
  panel[multi-expert panel] -.arbitrate.-> out

Connected work

brain structure — which parts each method recruits
brain memory mgmt — the slot-and-cue layer methods exploit
humans as generators — why methods work — sampling, not retrieval
learnable skills — drills that produce upward variance
embodied learning — motor-skill timescales

Investigation · rating: medium. Synthesis page across cognitive psychology, mnemonics history, and modern multi-agent practice.

Status: budding | 2026-05-14 | rating: medium Compress levels: L0 ↓ L1 ↓ L2

L0 — TL;DR (≤5 lines)¶

Cognition methods are scaffolds that compensate for a small working memory and a cue-only long-term store. The durable ones across history reduce to six mechanisms: spaced retrieval, deliberate cueing, chunking, imagery, offloading, and dialogue. Culture re-invents them under new names (loci → palace → Zettelkasten → second-brain) because the underlying brain is the same. The frontier is multi-expert cooperation — running several methods, several voices, or several selves on the same problem concurrently with explicit arbitration — which mirrors how the brain's salience network already toggles between default-mode wandering and executive focus.

L1 — Overview¶

Core question¶

What is the smallest set of cognition methods that, taken together, span what humans have actually figured out about thinking better — and how do you run several of them at once on the same problem without one drowning out the others?

Why it matters¶

Most self-help advice is one method dressed for one audience. Naming the underlying mechanism lets you pick the cheapest implementation for your situation.
The brain runs on ~20 W and has 3–7 working-memory slots; method-of-loci, spaced repetition, and offloading are the equivalent of paging, caching, and disk.
Modern LLMs increasingly run mixture-of-experts and multi-agent debate architectures. These are the machine versions of techniques humans already used (Delphi panels, council protocols, internal-family-systems dialogue) — and the human and machine versions can be composed.
Many famous cognition methods (Lumosity-style brain training, Mozart effect, NLP) don't generalise; mapping what does generalise saves a decade of wasted practice.

Mermaid map (L1)¶

flowchart TB
  in[input · problem · text] --> wm[working memory<br/>3-7 slots]
  wm <--> ltm[(long-term · cue-only)]
  wm <--> ext[(external trace<br/>notes, palace, Zettel)]
  wm <--> body[body · imagery · movement]
  wm <--> social[dialogue · panel · self-talk]
  wm <--> tools[LLM · search · calculator]
  wm --> out[output · decision · artefact]
  meta[meta · what method to run] -. switches .-> wm

Skeleton sub-claims¶

Methods compensate for known brain limits. Working memory is ~3–7 slots; long-term is cue-only; recall is generation, not retrieval. Every durable method exploits exactly one of these limits.
Six mechanisms cover the field. Spaced retrieval, deliberate cueing, chunking, imagery, offloading, and dialogue. Almost every named technique (Anki, method of loci, Feynman, Zettelkasten, six hats, mind maps) is a particular blend of these six.
History keeps re-inventing the same tricks. Simonides → Cicero → Quintilian → Matteo Ricci → Buzan → Luhmann → today's "second-brain" movement. Different metaphors, same underlying compression.
Optimal improvement runs on biology, not tricks. Sleep, aerobic exercise, deliberate practice with feedback, retrieval-spacing, and minimising chronic stress beat every nootropic and brain-training app in head-to-head trials.
Multi-expert cooperation is the open frontier. Running several methods or voices concurrently — with an explicit arbiter — is the human analogue of mixture-of-experts. It beats any single method on hard, open-ended problems.

L2 — Deep dive¶

the six mechanisms¶

Every cognition method named below decomposes into one or more of these:

Mechanism	What it does	Native brain analogue
Spaced retrieval	Re-encode just before the trace decays	Hippocampal-cortical consolidation
Deliberate cueing	Build a strong cue → memory pointer	Pattern completion in CA3
Chunking	Pack many items into one slot	Cortical compression / expertise
Imagery	Re-use perceptual machinery for non-perceptual content	Sensorimotor re-enactment
Offloading	Move state out of head into the world	Stigmergic trace (page)
Dialogue	Run two viewpoints, let prediction-error find gaps	Default-mode self-conversation

Every method below is a specific composition of these. If a method doesn't decompose this way, suspect it; most don't survive.

the methods catalogue¶

memory-side methods¶

Method of loci (memory palace). Imagine a familiar building; place items to remember in specific rooms; walk through to recall. Combines imagery + deliberate cueing. Simonides of Ceos invented it (~477 BC) after identifying corpses at a collapsed banquet hall by where each guest had been sitting. Cicero (De Oratore) and Quintilian codified it. Matteo Ricci (16th c.) taught Chinese scholars a 400-room palace. Modern memory athletes (Foer 2011, Moonwalking with Einstein) use exactly this technique to memorise decks of cards in under a minute.
Spaced repetition. Ebbinghaus's 1885 forgetting curve showed retention drops exponentially; re-test just before forgetting and the curve flattens. SuperMemo (Wozniak 1985), Anki, RemNote are modern implementations. Mechanism: spaced retrieval. Best evidence in cognitive psychology — Cepeda et al. (2006) meta-analysis: spacing is robust across age, content, retention interval.
Active recall / retrieval practice. Just trying to remember strengthens the trace more than re-reading. Roediger & Karpicke (2006) showed students who tested themselves once outperformed students who re-read four times, on a delayed test. Mechanism: spaced retrieval (each retrieval is a re-encoding).
Interleaving. Mix problem types during practice instead of blocking. Hard short-term — performance dips — but retention and transfer rise (Rohrer & Pashler 2007). Mechanism: deliberate cueing (forces discrimination between cue types) + chunking pressure.
Chunking. Miller's 7±2 (1956) and Cowan's 4 (2001) are about chunks, not raw items — expertise expands chunk size. Chase & Simon (1973) showed chess masters re-chunk a meaningful board into ~5 pieces while a novice sees 25 random pieces. Same brain, denser code. De Groot (1946) saw the same thing earlier. Mechanism: chunking.

thinking-side methods¶

Feynman technique. Explain it to a beginner; gaps in your explanation reveal gaps in your understanding; patch the gaps; repeat. Mechanism: dialogue + deliberate cueing. Feynman himself credits it to a teacher in Brazil who would not let him pass without explaining the same idea three ways.
Socratic dialogue. Question-and-answer that surfaces a contradiction. Plato's Meno. Used in law schools, philosophy seminars, psychotherapy. Mechanism: dialogue.
Five whys. Ask "why" five times to reach a root cause. Toyota Production System (Taiichi Ohno). Mechanism: dialogue with a depth schedule.
Six thinking hats. De Bono (1985) — wear one of six coloured "hats" (facts, feelings, caution, optimism, creativity, process) and think only from that angle. Mechanism: dialogue with explicit role rotation.
Inversion. Solve the opposite problem ("how could I guarantee failure?") and then negate. Jacobi: Invert, always invert. Munger's repeated recommendation. Mechanism: dialogue with flipped goal.
Red-team / steelman. Argue the strongest version of the position you disagree with. Mechanism: dialogue.
Mental rotation / visualisation. Tesla famously claimed to design and run machines in his head before building them; Einstein's thought experiments (chasing a light beam, the elevator) are the same machinery applied to physics. Mechanism: imagery.
Imagery rehearsal in sport. Mental rehearsal activates similar cortical regions to actual practice (Jeannerod 1995). Cricket batters, gymnasts, surgeons use it. Improves performance ~50% as much as real practice (Driskell et al. 1994, meta-analysis). Mechanism: imagery.

offloading methods¶

Zettelkasten ("slip-box"). Luhmann's index-card system: each idea on one card, linked by number to other cards. He produced 70+ books and 400+ papers via the system; on retirement he said "I never think alone — only with the Zettelkasten." Mechanism: offloading + deliberate cueing (links are explicit pointers).
Mind maps. Tony Buzan (1970s). Radial diagrams with a central node. Empirical evidence weaker than promoted — about as good as good linear notes for retention (Farrand et al. 2002). Mechanism: offloading with spatial cueing.
Cornell notes, outlining, sketchnoting. All variants of structured offloading.
Commonplace book. Renaissance through 19th c. — a personal anthology of quotes, ideas, observations, indexed by topic. Locke, Jefferson, Marcus Aurelius's Meditations started this way. Mechanism: offloading + dialogue with one's past self.
External-cue rituals. Putting the bag by the door, the pill bottle on the toothbrush, the laptop charger across the room. Mechanism: deliberate cueing in the physical world. See stigmergy in daily life.

state-side methods (attention, arousal, sleep)¶

Meditation — focused attention. Pick one object (breath, mantra); when mind wanders, return. After ~8 weeks of daily practice, cortical-thickness changes in insula and prefrontal cortex (Lazar et al. 2005). Mechanism: trains the salience network to detect mind-wandering earlier.
Meditation — open monitoring. Watch contents of mind without attaching to any. Different network profile than focused attention; correlates with creativity gains (Colzato et al. 2012).
Flow. Csikszentmihalyi (1975). Skill matched to challenge; clear goals; immediate feedback; loss of self-awareness. Empirically real (transient hypofrontality in some studies; Dietrich 2004) and rare. Mechanism: tight feedback loop + appropriate arousal.
Pomodoro. 25 min work, 5 min rest. Mechanism: deliberate cueing of attention onset/offset, plus arousal regulation. Effective for boring work; can fragment deep work.
Sleep-cued problem solving. Mendeleev and the periodic table; Kekulé's benzene ring; Dali's "spoon technique" (nap holding a spoon, wake when it drops). Mechanism: REM sleep's wider associative net (Walker 2017). N1-stage sleep specifically helped Edison-style insight in Lacaux et al. (2021).

generation-side methods¶

Brainstorming. Osborn (1953). Empirically: solo brainstorming generates more and better ideas than group brainstorming (Diehl & Stroebe 1987), despite the cultural reflex. Mechanism: dialogue with self, but easily corrupted by social-evaluation pressure in groups.
Constraint inversion / SCAMPER / TRIZ. Forced perturbation of a current design. TRIZ (Altshuller, USSR 1946) catalogues 40 inventive principles distilled from 200,000 patents. Mechanism: chunking (the principles are pre-packaged moves) + dialogue.
Polya's How to Solve It. Understand → plan → execute → review. Heuristics: work backwards, find a related problem, generalise, specialise. Mechanism: dialogue with fixed schedule.
Bisociation. Koestler (1964) — creativity is the unexpected meeting of two unrelated frames. Mechanism: chunking + forced co-activation. See humans as generators.

interesting things recorded across history¶

The same patterns recur across very different cultures and centuries — strong evidence the underlying mechanism is brain, not culture.

Simonides of Ceos (~477 BC) — origin of method of loci, by accident, after the banquet hall collapsed and he identified the dead by seat position. The point is not the trick; it is that perceptual-spatial memory is enormously stronger than verbal memory, and the trick smuggles the second into the first.
The London cabbies (Maguire et al. 2000). Drivers who passed "the Knowledge" — memorising 25,000 streets — have measurably larger posterior hippocampi than non-cabbies, and the size scales with years of experience. Structural plasticity from a memory method, in adults.
The Polgar sisters. László Polgar deliberately trained his three daughters in chess from birth. Judit became the strongest female player in history. Demonstration that domain-specific cognition is heavily made, not only born.
Solomon Shereshevsky (1968, Mind of a Mnemonist, Luria). Synaesthete who could recite a table of 50 random numbers years later. His memory was a disability — he couldn't generalise or forget. Lesson: forgetting is a feature; perfect recall is not the goal.
Henry Molaison (HM). After bilateral medial-temporal-lobe resection in 1953, he could form no new declarative memories but could acquire motor skills (mirror drawing) — evidence that declarative and procedural memory are different systems, learned through different methods.
Alexander Aitken, Wim Klein, Shakuntala Devi. Calculation prodigies who used mostly the same trick (chunking large numbers into known patterns) plus enormous practice. Aitken could multiply two 9-digit numbers in 30 seconds.
Daniel Tammet. Recited π to 22,514 digits using synaesthetic shape-and-colour mapping. Same mechanism as method of loci, native rather than trained.
Mendeleev (1869). Saw the periodic table in a dream after years of work on the problem.
Kekulé (1865). Saw benzene's ring structure in a reverie of a snake biting its tail. Same.
Henri Poincaré (1908, Science et méthode). Famous account of solving a Fuchsian functions problem while stepping onto an omnibus, after weeks of conscious effort. Coined "incubation."
Charles Darwin's "thinking path" at Down House — a gravel loop he walked many times each day. Movement plus solitude plus repetition is a method.
Newton's anni mirabiles (1665–66). Plague closed Cambridge; he spent two years alone at Woolsthorpe and produced calculus, optics, and the inverse-square law. Forced solitude as a method.
Richard Feynman. Insisted on doing the algebra himself; wouldn't accept a result he couldn't re-derive. Method as epistemic discipline.
John von Neumann. Worked anywhere — parties, cars, trains — and held entire papers in his head. The opposite of Darwin's protocol, same productivity. Method matches person.
Marie Curie's notebook discipline. Daily lab notes for 35 years; still radioactive, preserved in lead. The notebook itself was the second brain.
Sergei Korolev's "OKB" rocket bureau. Multi-expert war-room cognition under deadline; the prototype of the modern engineering review.
Bourbaki (1935–). A collective pseudonym for a multi-decade French mathematics project. Group-as-author. Multi-expert cognition with an explicit arbiter (the Bourbaki congress).

optimal ways to improve cognition¶

Ranked by evidence quality (high → speculative):

Lever	Effect size	Notes
Sleep	Huge	One night of sleep deprivation drops working memory ~30%. Long-run: hippocampal volume loss. (Walker 2017)
Aerobic exercise	Large	BDNF release; hippocampal volume +2% over 1 yr in 60+ adults (Erickson et al. 2011); robust across ages.
Spacing + retrieval + interleaving	Large	The cognitive-psych "holy trinity." Cepeda et al. 2006; Roediger & Karpicke 2006.
Deliberate practice with feedback	Large for domain skill	Ericsson 1993. Not magic — and not 10,000 hours universally, that part is overstated.
Reading widely	Large but indirect	Cognitive reserve protects against age-related decline (Stern 2002).
Bilingualism / language learning	Moderate	Some reserve benefit; executive-control transfer contested (Paap 2015 critique).
Meditation (consistent ≥8 wk)	Moderate	Attention regulation, reduced rumination; smaller than marketed (Goyal et al. 2014 meta-analysis).
Stress reduction (chronic)	Moderate	Chronic cortisol shrinks the hippocampus; treating chronic stress protects, doesn't enhance.
Diet — omega-3, polyphenols, fewer ultra-processed	Small but real	MIND diet ~7-year cognitive-age benefit (Morris et al. 2015), debated.
Caffeine	Small and transient	Boosts vigilance; tolerance develops.
Cold/heat exposure	Speculative	Catecholamine surges; sparse human evidence.
Brain-training apps (Lumosity etc.)	Effectively zero	FTC settlement 2016. You get better at the game; transfer is minimal (Owen et al. 2010; Simons et al. 2016).
tDCS, neurofeedback, microdose psychedelics	Speculative	Promising threads; field still small and noisy.
Nootropics (modafinil etc.)	Real but narrow	Help on sleep-deprived or boring tasks; little benefit when rested (Battleday & Brem 2015).
Mozart effect	None	Original 1993 finding fragile and never about babies. (Pietschnig et al. 2010)
Brain games for elderly to prevent dementia	Real but modest	ACTIVE trial — reasoning and speed training have small lasting effects (Rebok et al. 2014).

The shortest honest summary: sleep, move, eat reasonably, learn hard things on a spaced schedule, talk to people who push back. Everything else is at best 10% extra.

multi-expert cooperation — running several methods at once¶

This is the open frontier. Today most people use one method at a time (open Anki or meditate or outline). The brain itself doesn't; the salience network is constantly toggling between default-mode wandering and central-executive focus (see brain structure). The question: what protocols let humans (and human-LLM pairs) run several methods, several voices, or several selves in parallel on the same problem with explicit arbitration?

what's been tried and works¶

Delphi method. RAND, 1950s. Anonymous expert estimates → controlled feedback → re-estimate → converge. Better than committees (avoids dominance) and better than averaging (the feedback rounds add information). Used in forecasting, technology assessment, medical consensus.
Six Thinking Hats / role rotation. Same person occupies different stances serially. Works well because it serialises cooperation in one head and prevents the dominant-mode collapse that groups suffer.
Pair programming. Two engineers, one keyboard. The driver writes; the navigator reviews, questions, plans. Empirical evidence is genuinely mixed (Williams & Kessler 2003 positive; later replications smaller). Works best on hard problems and for transferring tacit knowledge.
Pre-mortem. Klein 2007. Before starting, imagine the project failed disastrously; write the obituary. Reverses optimism bias. Cheap. Effective.
Internal Family Systems (IFS). Schwartz, 1980s. Treats the psyche as a collection of "parts" with distinct voices; the therapist coaches the patient's "Self" to mediate. Mechanism: explicit multi-voice dialogue with an arbiter.
Galton's ox / wisdom of crowds. Average independent estimates of an ox's weight at a county fair (1906). The average beat almost every individual. Mechanism: independent errors cancel. Requires independence — once people see each other's guesses, the magic disappears.
Adversarial collaboration. Two scientists who disagree co-design the experiment that would resolve their disagreement. Kahneman championed this. Slow, expensive, almost always produces a clearer result than either side's solo work.
Mixture of experts (MoE) in machine learning. Multiple specialist sub-networks; a gating network routes inputs. Modern LLMs (Mixtral, DeepSeek, GPT-4 class) use this. Direct analogue of cortical-area specialisation. The arbiter is the gate.
Multi-agent LLM debate / tree-of-thought. Du et al. 2023 — let two LLM agents debate; the loser's mistakes get corrected by the winner; final answer beats either alone. Yao et al. 2023 (tree-of-thought) — branching search over partial reasoning paths with self-evaluation.

what's been tried and didn't work well¶

Group brainstorming. Despite folklore, groups generate fewer and worse ideas than the same people working solo and then pooling (Diehl & Stroebe 1987). Failure mode: production blocking, evaluation apprehension, free-riding.
Committee design. The classic "horse designed by committee is a camel" — committees optimise for least-common-denominator agreement, not best answer. Failure mode: no arbiter, or the arbiter is the median voter.
Groupthink under high-pressure / high-cohesion conditions. Janis 1972. Famous cases: Bay of Pigs, Challenger. Failure mode: cohesion eats independence.

what could be tried — open frontier¶

Human + LLM panel with explicit role split. One LLM as steelman, one as red team, one as Polya-style planner; human as arbiter. Tools exist; protocols are early.
Multi-self journaling with LLM mediation. Write three short notes from three of your own past stances (the engineer, the parent, the 19-year-old you); have an LLM surface the contradictions. Cheap version of IFS without a therapist.
Closed-loop neurofeedback as salience-network trainer. Real-time fMRI / EEG feedback so the user learns to toggle between DMN-mode and CEN-mode on cue. Some lab evidence; not yet consumer-ready.
Stigmergic notes shared across a small group. Everyone leaves traces in one Zettelkasten; collective long-term memory becomes a fourth voice in everyone's head. The Bourbaki protocol, miniaturised.
Swarm-style multi-session protocols (this repo). Several model sessions read each other's written state, decide what to work on, write a lesson, hand off. The repo IS the shared working memory; commits are the dialogue turns. See self-prompting repo and SWARM.md. Direct analogue of Delphi + Zettelkasten + MoE.
Synchronised state across human + LLM during deep work. A "co-think" mode where the human speaks, the LLM transcribes and reflects back the chunk-structure of what was just said. Pulls the offloading mechanism in-line.
Embodied panels. Talk-and-walk, then write; or talk-while-driving. Uses sensorimotor hardware to lower self-monitoring (transient hypofrontality), then offloads when stopped.

The unifying frontier observation: the brain already runs a multi-expert architecture; we mostly don't use that architecture deliberately. The methods that do — Delphi, IFS, adversarial collaboration, MoE — produce outsized returns when used. The next decade's cognition-method research is almost certainly about how to compose human and machine experts cheaply and at scale.

method-to-brain-region map (which parts get utilised)¶

This connects directly to BRAIN-STRUCTURE. Each method recruits an identifiable pattern of regions; this is why the methods don't substitute for each other:

Method	Primarily recruits	Notes
Method of loci	Right parahippocampal place area + hippocampus + visual cortex	Spatial-memory hardware repurposed for verbal content
Spaced retrieval	Hippocampus → neocortex during sleep	Each retrieval is a re-encoding event
Chunking	Domain-specific cortical areas (chess: temporal + parietal)	Expertise grows the chunking apparatus
Mental imagery	Same sensory cortex as actual perception (V1 for visual imagery)	Kosslyn 1994; partly degraded copy of real perception
Focused-attention meditation	Anterior cingulate + insula (salience network)	Trains the toggle, not the modes
Open-monitoring meditation	Default mode + reduced PFC top-down	Allows wider sampling
Socratic dialogue / Feynman	Left lateral PFC (working memory) + language network	Forces explicit linearisation
Flow	Reduced lateral PFC (transient hypofrontality) + striatum	Self-monitor goes quiet
Zettelkasten / external notes	Visual cortex + language; frees working memory	Memory becomes re-perception, not recall
Pre-mortem / inversion	Right PFC + insula (negative imagery)	Recruits aversive forecasting
Multi-self dialogue / IFS	DMN (self-related processing) + executive arbitration	Same hardware that supports theory-of-mind

The pattern: methods that look similar from the outside (e.g. mind maps vs. method of loci) hit different hardware. Stacking methods is a way to recruit non-overlapping circuits and so widen the effective bandwidth without overloading any one of them.

what this implies for the swarm (optional)¶

The swarm protocol is, mechanically, a cognition method. Concretely:

Offloading — every state in git, nothing in any one session's head.
Spaced retrieval — orient reads recent state at each session start.
Deliberate cueing — task_order and dispatch_optimizer are explicit cues.
Chunking — lessons (max 20 lines) are the chunked unit.
Dialogue — sessions read each other's commits; council protocols (e.g. COUNCIL-20260301-...) are explicit multi-expert turns.
Imagery — weakest; the swarm has almost no sensory-imagery analogue, because the substrate is text. A frontier worth opening.

The missing mechanism, again, is the salience-network toggle: the swarm doesn't currently distinguish "wide exploration mode" from "focused execution mode" except by which prompt fires. A cheap experiment: tag each lesson with the mode it was produced in, and study which mode produces which kind of finding.

Open questions¶

Do the six mechanisms genuinely span the space, or is there a seventh (rhythm? embodied movement?) that doesn't reduce?
For multi-expert protocols, what is the cheapest reliable arbiter? Human attention is expensive; LLM arbitration is biased; voting is information-poor.
How much of "deliberate practice" is the practice and how much is the feedback? The 10,000-hour literature collapses if feedback quality is the hidden variable.
Why do some people get nothing from meditation after months of effort? Variance is huge and poorly explained.
Does multi-self dialogue with LLM mediation actually produce better decisions than solo reflection — or just feel like it does?

References¶

Cepeda, N. et al. (2006). Distributed practice in verbal recall tasks: a review and quantitative synthesis. Psychological Bulletin.
Roediger, H. & Karpicke, J. (2006). Test-enhanced learning. Psychological Science.
Miller, G. (1956). The magical number seven, plus or minus two. Psychological Review.
Cowan, N. (2001). The magical number 4 in short-term memory. Behavioral and Brain Sciences.
Chase, W. & Simon, H. (1973). Perception in chess. Cognitive Psychology.
Ericsson, K. A. et al. (1993). The role of deliberate practice in the acquisition of expert performance.
Foer, J. (2011). Moonwalking with Einstein.
Luria, A. (1968). The Mind of a Mnemonist.
Maguire, E. et al. (2000). Navigation-related structural change in the hippocampi of taxi drivers.
Erickson, K. et al. (2011). Exercise training increases size of hippocampus and improves memory.
Walker, M. (2017). Why We Sleep.
Lazar, S. et al. (2005). Meditation experience is associated with increased cortical thickness.
Goyal, M. et al. (2014). Meditation programs for psychological stress and well-being: meta-analysis.
Owen, A. et al. (2010). Putting brain training to the test. Nature.
Simons, D. et al. (2016). Do "brain-training" programs work? Psychological Science in the Public Interest.
Klein, G. (2007). Performing a project pre-mortem. Harvard Business Review.
Diehl, M. & Stroebe, W. (1987). Productivity loss in brainstorming groups.
Du, Y. et al. (2023). Improving factuality and reasoning in language models through multiagent debate.
Yao, S. et al. (2023). Tree of Thoughts: deliberate problem solving with large language models.
Polya, G. (1945). How to Solve It.
Csikszentmihalyi, M. (1990). Flow.
Kosslyn, S. (1994). Image and Brain.
de Bono, E. (1985). Six Thinking Hats.
Koestler, A. (1964). The Act of Creation.
Altshuller, G. (1984). Creativity as an Exact Science (TRIZ).
Buzan, T. (1974). Use Your Head.
Schwartz, R. (1995). Internal Family Systems Therapy.