758+ definitions for the age of incentive intelligence.
Every bias, fallacy, mental model, neuroscience principle, perverse incentive, and AI alignment term we teach — each with a real-world example, why it matters in the executive room, the AI implication, and a counter-move. Linkable, citable, searchable.
How will I feel about this in 10 minutes / 10 months / 10 years?
A group decides on a course of action that nobody actually wants, because everyone assumes others prefer it.
Inattention or forgetfulness caused by low attention, hyperfocus, or distraction.
Deriving general rules from specific examples; the leap from instance to concept.
Written rules about how AI may be used internally.
Earn-outs designed to retain founders often demotivate the team they bought.
Doing something feels safer than doing nothing — even when nothing wins.
The minimum energy needed to start a reaction; same idea for starting habits or initiatives.
Power tends to corrupt, and absolute power corrupts absolutely.
We attribute our own actions to situations but others' actions to their character.
Attacking the person rather than the argument.
The brain evolved to reason adaptively, not always truthfully, to reduce the cost of errors.
We solve problems by adding, even when subtracting would be better.
Tenure-track jobs replaced by low-paid adjuncts, lowering cost and quality.
Innovators → early adopters → majority → laggards, AI-specific.
Holding a different — and correct — set of tools than everyone else.
Feelings act as shortcuts for facts.
Naming an emotion reduces its intensity.
Predicting how we will feel in the future, usually inaccurately.
We favor people who are similar to us or who like us.
Assuming that if one option is true, another must be false, when both can be true.
If A then B; B happened; therefore A.
AI system that takes actions to achieve goals, often across tools.
Presuming a purposeful actor behind events that may have no actor at all.
Someone who incites others to take illegal or rash action, often to expose or discredit them.
When the agent acts, who's responsible?
Autonomous agents deployed before liability frameworks exist.
Augmentation strategy vs. substitution strategy.
Logging of AI inputs, outputs, and decisions.
Inventory of models, data, tools, and dependencies in an AI system.
Tracing AI components for risk and compliance.
Cross-functional governance body for AI decisions.
Agents deployed before anyone owns the consequences.
Stages of organizational AI capability.
Individual productivity gains hide collective output degradation.
AI investment outpacing measurable productivity gains.
Categorizing AI use cases by risk level.
AI strategy = decisions about which capabilities to build and where.
Scarce AI talent commands market-distorting compensation.
Designing processes from scratch around AI capability.
AI generates content; AI scrapes content; AI trains on its own output.
Teams built around AI from day one vs. teams adding it to existing workflows.
What cannot be settled by experiment is not worth debating.
Discounting algorithmic advice even when superior.
A finite set of well-defined instructions for solving a problem or performing a computation.
Every model simplifies reality; some are still useful.
Researchers favor conclusions aligned with their school, team, or sponsor.
Cumulative physiological wear from chronic stress.
Combining substances to create a new material stronger than its parts.
Forgetting to compare an offer with the next-best alternative.
Outcomes that could have happened but did not.
We prefer known risks to unknown ones, even when the unknown is better.
Preferring options with known probabilities over options with unknown ones.
Emotional override of rational processing.
Overthinking a situation so that decision-making stalls.
Over-reliance on the first number that hits the table.
A single story used as proof of a general claim.
Using personal stories or isolated examples instead of evidence.
Annual bonuses reward annual results — risk that blows up in year four with damage in year seven is rational.
Once-a-year feedback rewards once-a-year behavior.
Brain region tracking conflict, error, and effort.
Treating AI as more humanlike than it is.
We treat AI as more humanlike than it is.
Systems that gain from disorder.
Perceiving meaningful connections in unrelated things.
It's true because authority says so.
Common sense says so, therefore it's true.
Believing something is true because of the consequences of its truth.
Substituting feeling for argument.
Using fear instead of evidence.
Claiming something is true because it hasn't been proven false (or vice versa).
Claiming something is true or better because most people believe it.
If it's 'natural,' it's good.
It's better because it's newer.
Asking for a conclusion based on sympathy rather than logic.
Assuming something is true because it is probable or possible.
Rejecting an argument because of dislike for the source or beneficiary.
It's right because it's how we've always done it.
Exploiting price differences between markets for risk-free profit.
Using an authority's opinion as evidence, regardless of its merits.
Concluding that a claim is false because the argument for it is flawed.
Reasoning backward from outcomes to motives.
Claiming something is true because it has not been proven false.
I can't believe it; therefore it isn't true.
n=1 generalized to n=everyone.
Saying it often enough until objections fade.
Saying it enough times until it sounds true.
Drawing a conclusion from the absence of evidence rather than its presence.
Sun Tzu's ancient treatise on strategy, deception, and positioning.
Assuming qualities of one thing transfer to another because they are associated.
Masking the sponsors of a message to make it appear grassroots.
Conflict between belligerents with very different resources and tactics.
Attention is the most contested resource in modern work.
Task switches leave cognitive residue that degrades performance.
Our perception is shaped by what we selectively pay attention to.
Disagreement grows more extreme as the parties think more about the issue.
Replacing a hard question with an easier one without realizing it.
Systematic errors in explaining the causes of behavior — yours or others'.
Auditors paid by the firms they audit have predictable blind spots.
Augment when judgment matters. Automate when scale matters.
We overweight whoever has the bigger title in the room.
Favoring suggestions from automated systems over conflicting human judgment.
Reduced vigilance with automated systems.
Self-direction is one of the deepest motivational levers.
Relying on examples that come to mind easily, not on actual frequency.
A story repeats until it becomes obviously true.
We judge frequency by how easily examples come to mind.
Contradictory evidence sometimes deepens the original belief.
Adopting beliefs because the room already did.
It's true because many believe it.
Combine extreme safety with extreme risk; avoid the middle.
We accept vague, general statements as personally meaningful.
Ignoring general statistics in favor of specific, vivid information.
We ignore underlying probabilities in favor of vivid specifics.
Posterior = (likelihood × prior) / evidence.
Update beliefs in proportion to the strength of new evidence.
Start with a prior; update with new evidence.
Adding the word 'because' (with any reason) increases compliance dramatically.
Assuming the conclusion in the premise.
Novices experiencing early success, often due to variance and small samples.
Approach a situation as if you knew nothing.
Willpower fails; structure works.
Behaviors spread through groups via observation and imitation.
Evaluating an argument's logic based on whether you agree with the conclusion.
Beliefs survive the evidence that should have killed them.
We come to like people we've done a favor for, more than people who've helped us.
Passion is inversely proportional to the amount of real information available.
Any headline that ends in a question mark can be answered by the word no.
A color appears different depending on adjacent colors.
We see bias in others more easily than in ourselves.
Systematic skew in model behavior across groups.
Spending disproportionate energy on trivial decisions.
Subconscious preference for numbers (and letters) that appear in our own name or birthday.
Unusual or bizarre material is more memorable than common material.
High-impact, hard-to-predict, retrospectively explainable events.
Quarterly bonuses create end-of-quarter behavior changes.
A persuasive message produces the opposite of its intended effect.
Re-hires often get raises larger than internal promotions.
Boredom triggers creative wandering.
A single constraint limits the throughput of an entire system.
A system's throughput is constrained by its single slowest step.
Remembering more context around a scene than was actually present.
Decisions are constrained by available information, cognitive limits, and time.
The mind can be socially influenced toward conclusions that bypass scrutiny.
The energy needed to refute bullshit is an order of magnitude bigger than to produce it.
Offering value in exchange for influence over a decision.
Small visible disorders signal that bigger ones will be tolerated.
Adding manpower to a late software project makes it later.
Adding people to a late project makes it later.
Strategic choice on AI capability sourcing.
Assuming the opponent is wrong, then explaining why they came to that wrong belief.
Asking the other side to disprove an unsupported claim.
Diffused responsibility means nothing happens.
The more a quantitative indicator drives decisions, the more it distorts the process it measures.
AI capability outpacing organizational ability to use it.
AI bolted onto existing workflows to look forward-leaning.
A substance that speeds up a reaction without being consumed.
Using a fake identity to deceive someone, usually for gain or manipulation.
Pay tied to stock price encourages short-term price management.
Nonlinear systems are highly sensitive to initial conditions.
Deliberate, sustained effort to damage someone's reputation.
Donors penalize 'overhead'; charities under-invest in capacity.
Knowledge from someone who can recite the answers but doesn't understand them.
People look better in groups than as individuals.
Every unnecessary element in a story must be removed, or it will be expected to matter.
Inputs combine under conditions to produce new outputs — sometimes irreversibly.
Quoting only the studies that support the position.
Selecting only the data that supports the conclusion.
Don't remove something until you understand why it was put there.
Adults can't recall early childhood; the memories never consolidated.
The threat of retaliation suppresses legitimate speech or action.
How options are presented changes which ones get chosen.
Too many options produces worse decisions and less satisfaction.
We remember our past choices as better than they were.
Cognitive performance varies predictably with circadian rhythm.
Temporal illusion where the first moment after a fast event seems stretched.
Grouping information into familiar units to improve memory and processing.
The rate at which customers (or employees) leave over a period.
Higher rates of mistreatment toward stepchildren than biological children.
Know the perimeter of what you actually know.
The conclusion is hidden inside the premise.
Sufficiently advanced technology is indistinguishable from magic.
Associating a neutral stimulus with a meaningful one until the neutral one triggers the response.
Headlines optimized for clicks become editorial policy.
Seeing patterns in random data.
With clear property rights and low transaction costs, parties negotiate to efficient outcomes.
A reward designed to reduce X produces more X.
The ability to focus on one voice in a noisy environment.
Systematic deviations from rationality in judgment.
Most biases are efficient heuristics misapplied.
Psychological stress from holding contradictory beliefs; usually resolved by changing the easier one.
Working memory has limits.
Working memory has limits.
Outsourcing thinking to tools shifts what we remember and learn.
Pausing System 1 to engage System 2.
Striking pattern that is statistically expected in large samples.
Once we say it publicly, we'll defend it past the data.
A mechanism that locks you into a future action to overcome present temptation.
Information that everyone knows, and everyone knows that everyone knows.
Treating multiple data points as independent when they came from one source.
Specialize in what you give up the least to do.
We care less as the number of victims increases.
Complements raise each other's value; substitutes lower it.
A system with many interacting parts that learn and adapt.
Many interacting parts producing emergent behavior nobody designed.
Compliance programs designed to satisfy regulators, not prevent harm.
What's true of the parts is assumed true of the whole.
Small advantages, applied consistently, become uncatchable.
GPU access and pricing shape what's feasible.
The relationship between inputs and outputs changes.
Producing fabricated memories or explanations without intent to deceive.
Quantifying uncertainty in model outputs.
Exploiting trust to defraud — only works because the mark wants to believe.
We hunt for evidence that we were right all along.
A hidden variable that influences both supposed cause and supposed effect.
Testing hypotheses only by looking for confirming evidence.
Believing a specific scenario is more likely than its more general one.
Updating beliefs too slowly when new evidence arrives.
Remembering past attitudes and behavior as more consistent with present ones.
Models trained to follow a written set of principles.
Avoiding contact with people or things perceived as 'contaminated'.
Cryptographic tracking of content origin.
Environmental factors influence perception and judgment.
How much input the model can process at once.
Removing a quote from its context to distort its meaning.
Believing previously learned misinformation even after it has been corrected.
Because there's no sharp line, no distinction exists.
Evaluations are warped by what came immediately before.
Convex payoffs gain more than they lose; concave do the opposite.
Software architecture mirrors org structure.
Renewed interest in a stimulus when a novel stimulus is introduced.
Coordinated action that produces outcomes unavailable to individual actors.
Working together has a cost that scales with the number of people.
Treating two things that move together as one causing the other.
Stress hormone with cognitive and immune-system consequences.
Operating economics shaped by per-token pricing.
Fixed costs don't scale with output; variable costs do.
Testing model behavior on hypothetical alternate inputs.
Imagining what could have happened instead of what did.
Winning the disputed population, not just defeating the visible opponent.
Countering hostile narratives with truthful, well-timed alternatives.
Giving polite rather than honest answers to avoid offense.
Innovation destroys old industries and creates new ones.
A conditioned response changes disproportionately when reinforcement changes.
The threshold at which a system becomes self-sustaining.
We recognize faces from our own race more accurately than others.
The gap between early adopters and the early majority kills most products.
Groups sometimes behave as foolishly as the worst individuals.
Influencing the beliefs and actions of a crowd through psychological techniques.
Experiencing a remembered idea as if it were original.
Failure to recall when the retrieval cue is absent.
Judging phenomena by the standards of one's own culture.
With this, therefore because of this.
The best way to get a right answer online is to post a wrong one.
The drive to explore and understand for its own sake.
Forgetting what it's like not to know what you know.
Familiar information feels true.
Underlying data distribution changes over time.
Tracking where training and inference data came from.
Reducing cognitive biases through training, structure, and process.
Decision quality degrades over the course of a day.
Believing that society or one's institution is in inexorable decline.
Adding a clearly worse option steers people toward the option you wanted.
Whatever is pre-selected wins more often than it should.
Defaults bypass active cognition.
Brain network active during rest and mind-wandering — and during creative insight.
Unconscious psychological strategies that protect self-image.
Attributing causes to protect oneself from blame or vulnerability.
Decisions made to be defensible, not to be right.
Quietly redefining a term mid-argument.
Seeing every problem through the lens of one's profession.
Eroding the will to act, even when capacity remains.
Refusing to accept reality, despite evidence.
We spend less when holding large bills than small ones — same money, different behavior.
If A then B; not A; therefore not B.
Munger's principle: the surest way to get something is to deserve it.
Designing systems for the lowest-capability user produces resilience for everyone.
A designated dissenter in any major decision.
The idea that material conditions and contradictions drive historical change.
Glucose, omega-3s, sleep, and hydration each influence decision quality.
Adding noise to data to protect individual privacy.
Innovators → early adopters → early majority → late majority → laggards.
Incompetent employees are promoted to management to limit their workflow damage.
Each additional unit produces less marginal benefit.
Applying higher scrutiny to evidence we disagree with than to evidence we agree with.
Actively seeking evidence that would disprove your hypothesis.
How much we value the present over the future.
Deliberately false information spread to deceive.
We dismiss the ideas and merits of people we dislike.
Selling winners too early and holding losers too long.
Low-end or new-market entrants overtake established incumbents.
Two options seen side-by-side look more different than they are in isolation.
Multitasking measurably degrades performance.
Reasoning in distributions — ranges and probabilities — rather than points.
What's true of the whole is assumed true of the parts.
Insights from one domain don't automatically transfer to another.
Neurotransmitter of anticipation, not pleasure.
Once any player dopes, everyone is forced to consider it.
Resolving doubt by jumping to a conclusion — any conclusion — quickly.
Bad teams 'tank' for better draft positions.
Dread weighs roughly double in our calculus what the equivalent gain does.
Substance dependence distorts judgment and accelerates decline.
Two ways of thinking, both real, both useful, often mis-deployed.
~150 — the cognitive limit on stable social relationships.
Low ability paired with high confidence.
We remember peaks and ends, not the duration in between.
Inability to think rationally despite adequate intelligence.
Quarterly earnings drive quarterly behavior.
Drawing conclusions about individuals from group-level data.
Interdependent network of actors evolving together.
We value outcomes more when we worked hard for them.
Self-control draws from a limited daily pool. (Contested in lab; observed in life.)
Over-weighting one's own perspective when reconstructing events.
Prior solutions block better new ones.
Urgent vs. important; act, schedule, delegate, delete.
Important-but-not-urgent is where strategic leverage lives.
Constraining the option set to bias the decision.
Politicians optimize for the next election, not the next generation.
Vector representation of content for similarity and search.
Thinking happens through the body, not despite it.
The whole has properties that the individual parts do not.
Emotional states distort cognition and decision-making.
Emotions spread through groups via micro-cues.
Cold-state decisions don't survive hot-state moments.
Empathy: feel with. Compassion: act for. Different neural circuits.
What we wear influences how we think and perform.
We believe we will change less in the future than we actually will.
We value things more once they're ours.
The brain minimizes effort by default — most cognition is on cruise control.
Outrage outperforms accuracy on every engagement metric.
Time-on-app is the wrong thing to maximize for users — but the right thing for revenue.
Systems tend toward disorder unless energy is invested to maintain order.
We measure ourselves against peers, not against our past selves.
Systems maintain stability by self-regulating around a setpoint.
Two things may be equivalent in effect but not in form.
Using a word with two meanings as if it meant one.
Average outcomes across the population differ from outcomes across time for one person.
Continuing a failing course of action because of prior investment.
Insisting on a word's original meaning over its current one.
Comprehensive AI regulation in the EU.
Remembering past experiences as more positive than they were.
Evaluation suites that no longer reflect real-world conditions.
Systematic testing of model quality, safety, and capability.
Variation, selection, retention — the engine of cumulative change.
Predicting more extreme outcomes than actually occur.
Aerobic exercise measurably improves executive function.
Assuming something exists because we can name or define it.
Decisions optimize utility, not value.
Probability × payoff, summed across outcomes.
Why did it produce this? vs. How does it work?
Growth whose rate becomes ever more rapid in proportion to the growing quantity.
Ignoring the size of a sample or population in our judgments.
Costs or benefits that affect third parties not involved in a transaction.
Believing others are motivated by extrinsic rewards while we are motivated intrinsically.
Doing something for an external reward.
Negative emotions fade faster than positive ones.
Quantitative measures of model behavior across groups.
False information presented as legitimate journalism.
Errors in reasoning that undermine the logic of an argument.
Concluding that because an argument contained a fallacy, its conclusion must be false.
Reasoning that two things are alike based on shallow similarity.
Crediting a source or cause that isn't actually responsible.
Correlation does not imply causation.
Assuming that because B followed A, A caused B.
Assuming everyone secretly agrees with us.
Presenting two options as the only possibilities.
Presenting two options as the only possibilities when more exist.
Treating two things as comparable when they're not.
An act committed to look like it was done by someone else.
Recalling something that did not happen or recalling it differently.
Underestimating how common our positive traits and behaviors are.
For a claim to be scientific, it must be possible to prove it false.
As more facts are learned about a topic, retrieving any one fact slows.
A strategy to influence behavior by amplifying fear and uncertainty.
We notice what's present more than what's absent.
Training models across devices without centralizing data.
Paying providers per procedure rewards more procedures, not better outcomes.
Outputs of a system are routed back as inputs, amplifying or dampening change.
A rough calculation using order-of-magnitude reasoning.
Decompose a problem into estimable parts; multiply through.
Mental discomfort from holding conflicting beliefs or values.
Algorithms show us content that reinforces our existing views.
Specialized training on domain data.
Break a problem down to irreducible truths and rebuild from there.
Reasoning from foundational truths rather than from analogy or convention.
Mind latches onto the first plausible explanation and resists alternatives.
Asking 'why' repeatedly to trace a problem to its root cause.
Who, what, when, where, why — the minimum facts for any account.
Believing abilities are static and cannot be developed.
Neurological state of peak performance and engagement.
IQ scores have risen substantially over generations.
Self-reinforcing loops of momentum.
Two modes the brain alternates between.
We overweight one aspect of an event when predicting its impact.
Sales forecasts under-set to ensure attainment bonus.
We accept vague, positive personality descriptions as highly accurate.
A handful of providers shape the entire AI economy.
The same information lands differently depending on how it's wrapped.
Living systems minimize surprise to survive.
After noticing something, we start seeing it everywhere.
Seeing things only in their conventional use.
Their failure: character. Our failure: situation.
Research outcomes tend to favor the interests of the funders.
Stacking so many tenuous links that the conclusion sounds clever.
A complex system that works is invariably found to have evolved from a simple system.
Believing past random events influence future independent ones.
Decisions depend on what other strategic actors will do.
Making someone question their own reality or perceptions.
Experts believe news articles outside their field despite knowing their own field is misreported.
Judging an argument by its origin rather than its merit.
Burying the room in a flood of weak arguments faster than they can be addressed.
As online discussions grow longer, the probability of a Nazi comparison approaches one.
Low expectations produce low performance.
When a measure becomes a target, it ceases to be a good measure.
Optimizing a proxy of the goal degrades the actual goal.
Professors rewarded by student evaluations have an incentive to inflate.
Bad money drives out good when both are legally equivalent.
Groups make more extreme decisions than individuals would alone.
Desire for conformity leads to irrational or dysfunctional decisions.
Believing abilities can be developed.
Cue → routine → reward.
New habits anchored to existing ones stick faster.
Repeated stimuli produce diminishing response.
Confident outputs that are factually wrong.
Confident incorrect outputs may rank higher than hedged correct ones.
One strong trait colors judgment of unrelated traits.
Don't attribute to malice what's explainable by incompetence — or incentives.
Sweeping conclusions from a small sample.
People change behavior when they know they are being observed.
Managers measured by team size grow teams beyond need.
Neurons that fire together wire together.
Intersection of what you're best at, passionate about, and economically viable in.
Hedgehogs know one big thing; foxes know many small things.
We return to a baseline of well-being after gains or losses.
In quantum mechanics, certain pairs of properties cannot both be precisely known.
People adopt behaviors and opinions because others do.
Decision time grows with the number of choices.
Believing past events were obviously predictable once we know how they ended.
What can be asserted without evidence can be dismissed without evidence.
Everything takes longer than you expect, even when you account for Hofstadter's Law.
Systems should be understood as wholes, not just as collections of parts.
Believing streaks predict future streaks.
Human review at critical AI decision points.
Human oversight without per-decision review.
We disproportionately prefer rewards now over rewards later.
Symptoms or beliefs spread through a group with no underlying physical cause.
Threats to identity activate strong defensive responses.
We overvalue things we built ourselves.
Believing we influence outcomes that are largely random.
We overestimate how much others understand our thoughts and feelings.
Confidence in predictions outruns their actual accuracy.
Most of us think we're above average. Statistically, we can't be.
We overestimate the intensity and duration of emotional reactions.
Subjective feeling of not deserving one's success.
Doubt about one's abilities and fear of being exposed as a fraud.
Models learning from examples in the prompt.
We extend more trust, benefit of the doubt, and rope to 'our people.'
We miss unexpected events when focused on something else.
Training is a one-time cost. Inference is forever.
Authenticity becomes a performance the moment it becomes a paycheck.
One party has more or better information than another.
Believing more information leads to better decisions, regardless of relevance.
People copy the choices of others, assuming they have private information.
We favor members of our own group over outsiders.
Outer: the spec matches our intent. Inner: the model actually pursues the spec.
Awards reward visible novelty; quiet excellence goes unrecognized.
Innovation programs designed to signal innovation, not to produce it.
Doing everything right by current customers can kill you.
We ignore sample size when judging probability.
Different agents pursue the same subgoals regardless of their ultimate goals.
Brain region tracking internal state and gut feeling.
Insurers profit when claims are denied, delayed, or abandoned.
Managers hoard talent; cross-team mobility dies.
Decisions involve trade-offs between costs and benefits at different times.
Doing something because the doing is the reward.
Solve forward by asking how to fail.
Bureaucracies tend to expand and protect themselves, regardless of their original purpose.
Bypassing model safety constraints.
Redesigning roles around AI capability.
Customers hire products to do a job in their life.
Implying conclusions through leading questions while disclaiming responsibility.
Believing people get what they deserve.
Bet size optimized to maximize long-run growth without ruin.
Keep it simple, stupid. Simplicity beats complexity in design and communication.
Once we know something, it is hard to imagine not knowing it.
How much delay the user experience tolerates.
A diverse set of models from many disciplines, used together.
As sample size grows, sample averages converge to expected values.
We expect small samples to reflect population properties.
Organizations give disproportionate weight to trivial issues.
When trivial metrics become the target, the trivial becomes the strategy.
When a system in equilibrium is disturbed, it shifts to counteract the change.
Giving up after repeated exposure to uncontrollable negative events.
Performance improves with practice, but gains slow over time.
When quality is hard to verify, bad products drive out good ones.
We're more persuaded by people we like.
The longer an idea has survived, the longer it likely will.
Expecting change to happen at a constant rate.
Contingency-fee structures shape which cases get filed.
Words chosen to bias the audience toward a conclusion.
A question with a built-in unproven assumption.
Optimizing in a small region while missing a better global solution.
The best place nearby vs. the best place overall.
Customers become trapped in a product due to switching costs or network effects.
Aggregating many niche markets equals or exceeds the mainstream.
Neural mechanism of learning — connections strengthened by repeated activation.
Losses hurt roughly twice as much as equivalent gains feel good.
Loss processed more intensely than equivalent gain.
How outcomes are weighted in the decision calculus.
Treating real-world uncertainty like a game with clear, known rules.
Models are simplifications — never the thing itself.
Build buffers so small mistakes don't become fatal.
Decisions hinge on the next unit, not on the average.
Each additional unit of a good provides less satisfaction than the previous one.
Markets fail to allocate resources efficiently on their own.
When all you have is a hammer, everything looks like a nail.
The drive to get better at things that matter.
Relying solely on metrics that are easily quantified while ignoring what matters.
Memories are reconstructed each time we recall them, not retrieved intact.
We treat money differently depending on which bucket it's in.
Always have a working model of what your counterpart believes.
Familiarity breeds approval.
The trained model develops its own internal optimizer.
Combining results from multiple studies to find robust conclusions.
Thinking about one's own thinking.
Assuming the truth must lie between two extremes.
People will obey authority even when asked to harm others.
We want what others want.
Choose the option whose worst-case regret is least bad.
The simplest version of a product that can test a core hypothesis.
A consistent minority can shift the majority's views over time.
Neurons that fire both when acting and when observing the action.
Documentation of model purpose, performance, limitations, and risks.
Train a smaller model to imitate a larger one.
Performance degradation as real-world data shifts.
Insulation from risk changes the risks people take.
Past good behavior makes us feel entitled to later bad behavior.
Deriving 'is' from 'ought.'
Originators paid on volume, not on default rates.
Today's metrics achieved by quietly borrowing from tomorrow.
Changing success criteria whenever the prior criteria are met.
Multiple specialized agents coordinating on tasks.
Anything that can go wrong will go wrong.
We prefer stories over raw facts, even when the story is misleading.
A state where no player benefits from changing strategy unilaterally.
Deriving 'ought' from 'is.'
Bad lands harder than good.
Bad lands harder than good in the brain.
Value grows with the number of users.
The brain reshapes itself based on use.
U.S. voluntary AI risk management framework.
Redefining a category to exclude counterexamples.
A symmetric bell-shaped distribution where most values cluster around the mean.
A bias against ideas or products that originated outside the group.
Notifications calibrated for return visits, not value.
What the model is actually optimizing.
Prefer the simplest explanation consistent with the evidence.
When OKRs are tied to comp, ambition disappears.
Ambitious objectives paired with measurable key results.
We judge harm by action more harshly than the same harm by inaction.
Where the model runs shapes privacy, latency, cost, and capability.
Observe, Orient, Decide, Act — faster than your competitor.
Designed for collaboration; produces measurable productivity decline.
Open-weight vs. API-only models.
Behavior shaped by reinforcement and punishment over repeated trials.
The value of the next-best thing you could have done.
Underestimating the probability of bad outcomes — especially to us.
Investments that give you future choices.
Avoiding information that might be unpleasant.
We see 'them' as more uniform than 'us.'
Judging a decision by its outcome rather than by the quality of the process.
Confidence routinely outruns calibration.
A model that explains historical data too closely will fail on new data.
Adding external rewards reduces intrinsic motivation.
The range of policies or ideas that are politically acceptable at a given time.
Neurochemistry of trust, bonding, and in-group loyalty.
Most outcomes come from a few causes.
Work expands to fill the time available for it.
Where you can go is constrained by where you've been.
We judge experiences by their peak moment and how they ended.
I can't imagine how it works, therefore it doesn't.
Overestimating the probability of bad outcomes.
Worst-cases assumed as base-cases because they 'feel responsible.'
People are promoted until they reach a position where they are incompetent.
Drug companies optimize for high-margin chronic conditions, not cures.
Specialists earn more for procedures than primary care for prevention.
AI pilots that succeed and never scale.
We underestimate time and cost; we overestimate benefit.
Structuring actions so that responsibility can be denied.
The moment after which reversing a course becomes impossible or extremely costly.
Discrediting a source or person before they make their argument.
Subordinates report what leaders want to hear, not what's true.
We tend to remember pleasant information more accurately than unpleasant information.
Autonomic nervous system shapes safety, threat, and connection responses.
Short focused intervals + breaks outperform marathon sessions.
Legislators rewarded for delivering local benefits financed by national costs.
B followed A, therefore A caused B.
A relationship where a small number of inputs account for most outputs.
A small number of items account for most of the value.
Imagine the project failed; explain why.
The brain is a prediction engine; surprise is the signal to update.
Imagine the failure beforehand; investigate it afterward.
Today's discomfort outweighs tomorrow's reward.
Agents act in their own interest, not the principal's.
Individual rationality produces collective irrationality.
Assuming new technology will be adopted faster and more usefully than it will.
Reason in distributions, not in points.
We ignore probability when outcomes are emotionally charged.
Lowest-bid procurement produces lowest-quality outcomes.
Individual speed gains hide collective quality decline.
We assume our current state will persist into the future.
Promoting your best individual contributor to manager loses you both.
Titles substitute for raises; senior titles inflate, value doesn't.
Cache common prompt prefixes to reduce cost and latency.
Malicious instructions hidden in user input or retrieved content.
People value gains and losses asymmetrically, and judge relative to a reference point.
The shared belief that the team is safe for interpersonal risk-taking.
Meaning beyond the task multiplies motivation.
High expectations produce higher performance.
90-day reporting windows shape multi-year strategies.
Universities optimize for ranking metrics rather than education quality.
Tell someone they must, and they'd often rather not.
Overweighting whatever just happened.
When someone does us a favor, we feel pressure to return it.
Recommenders optimizing engagement produce radicalization as a byproduct.
Introducing an irrelevant topic to distract from the issue.
A group assigned to challenge plans and find weaknesses.
Adversarial review by an internal opposition.
Adversarial testing of AI systems.
Beliefs about reality shape the reality.
Extreme outcomes tend to be followed by more average ones.
Choose the option you'll regret least at 80.
Agencies meant to regulate an industry get captured by it.
Treating an abstraction as if it were a concrete thing.
We judge probability by how similar an event is to a prototype.
Pre-deployment analysis of who loses what.
Investing in workforce transition vs. workforce change.
Overestimating our ability to control future impulses.
Generate based on retrieved documents, not just trained weights.
Two-way doors vs. one-way doors.
Extreme results tend to be followed by less extreme ones.
Frequency and variability of reward shape behavior strongly.
Maximizing the reward signal in unintended ways.
The brain rewards the gap between expectation and outcome.
People take more risks when they feel safer.
Risk has known probabilities. Uncertainty doesn't.
Reinforcement learning from human feedback.
We remember the past as better than it was.
Return-to-office mandated for visibility, not productivity.
A quick way to estimate how long it takes an investment to double at a given growth rate.
Technologies grow slowly, then explode, then plateau.
Brain system that decides what's important right now.
Teams under-reporting AI capability to protect comp or status.
Choosing a good-enough option rather than optimizing for the best.
In any dispute, the intensity of feeling is inversely proportional to the value of the stakes.
Rare things feel more valuable, regardless of actual value.
Status, Certainty, Autonomy, Relatedness, Fairness — drivers of social brain response.
The default coordination outcome people converge on without communicating.
Projects gradually expand beyond their original goals.
Ask 'and then what?' — twice.
Your sample isn't random.
Humans need autonomy, competence, and relatedness.
Belief in one's ability to execute.
A belief that becomes true because people act as if it is true.
Wins are ours; losses are circumstantial.
Reflexively rejecting evidence that contradicts established norms.
Repeated stimuli sometimes produce stronger response.
Executives leave well; employees leave thin.
Employees using unauthorized AI tools to get work done.
EPS-targeted comp incentivizes buybacks even when reinvestment yields more.
What is considered normal changes gradually, making degradation invisible.
Institutions will try to preserve the problem to which they are the solution.
The ratio of useful information to irrelevant information.
A trend appears in different groups but disappears or reverses when groups are combined.
Assuming one cause where many are at play.
Single-loop fixes errors; double-loop questions the assumptions producing them.
Foundational skills erode through AI offloading.
Decision-makers should bear the consequences of their decisions.
Memory and skill consolidate during sleep.
A small step inevitably leads to extreme consequences.
Small step → catastrophic outcome, without the intermediate causal links.
Ideas, behaviors, and attitudes spread through social networks.
Social rejection activates the same brain regions as physical pain.
We look to others to decide what's correct.
Small changes that are not noticeable can add up to a significant change.
Spread learning over time for stronger retention.
Exempting a case from a rule for no principled reason.
The model achieves the goal as stated, not as intended.
Police evaluated on ticket counts produce more tickets, not safer roads.
Effects of an intervention spread to other areas beyond the intended target.
We assume others notice us more than they do.
Forced curve performance management produces internal sabotage.
Manager calibration produces predictable distortions over years.
Much of human behavior is positioning for status.
Preference for what already exists, simply because it exists.
Brain weights existing state heavily.
Threats to status trigger the same neural alarm as physical threats.
AI doesn't just replace tasks — it threatens identities.
Restate the opposing view at its strongest before responding.
Restating the strongest version of your opponent's argument before responding.
Stocks are accumulations; flows are rates of change that affect them.
Causal chain from inputs to outputs to outcomes to impact.
Misrepresenting an argument to make it easier to attack.
Looking where the light is, not where the keys are.
Have a thesis. Update it on evidence.
Easy to start, hard to leave.
Past investment should not influence future decisions.
Past investment is irrelevant to future decisions.
Throwing more in because we already threw a lot in.
Studying winners while ignoring identical losers.
Generalizing from winners while ignoring identical losers.
Strategy copied from survivors ignores identical strategies that died.
Accidents occur when multiple layers of defenses fail simultaneously.
What does it cost to leave?
Generated training data that mimics real data.
Fast intuitive vs. slow deliberative thinking.
We defend existing systems even when they harm us.
Behavior emerges from structure, not from individuals.
Knowledge that is hard to transfer because it is personal and experience-based.
Goals become the focus, sometimes at the expense of the underlying purpose.
Schools evaluated on test scores teach to the test.
Drawing a target around wherever the bullets landed.
Cherry-picking data to fit a pattern after the fact.
Modeling what others think, want, and will do.
Modeling what others think, feel, and intend.
A neutral account of a conflict that both sides can agree on.
A phrase that ends thought rather than advancing it.
Under threat, the prefrontal cortex goes offline.
Our decisions are influenced by when events occur or when we evaluate them.
Start cooperative; copy your counterpart's last move thereafter.
Models calling external functions and APIs.
Real cost includes data, ops, monitoring, governance, training.
Shared resources get over-consumed by individual rational actors.
Coaches paid on titles take more risk; coaches paid on attendance take less.
Matching trust in a system to its actual reliability.
Deflecting by pointing to the other side doing it too.
Justifying a wrong because someone else also did wrong.
Teams small enough to be fed by two pizzas.
New hires paid more than tenure; tenure responds rationally.
False positives vs. false negatives.
A small group can dominate when the majority is silent or divided.
Negative behavior of outgroup members is attributed to character; positive behavior to situational factors.
Binding yourself now to avoid future temptation.
Actions have effects that were not planned or anticipated.
Metrics that look good without indicating business value.
Database optimized for high-dimensional similarity search.
Design rules without knowing where you'll land under them.
Concentration risk on a single AI provider.
Where, how, what — in that order.
Items that stand out are more likely to be remembered.
Concentrated benefits + diffused costs = the few win against the many.
Trust deepens when people can admit weakness without penalty.
Reporting wrongdoing is personally costly; staying silent is personally rational.
Aggregated independent judgments can be more accurate than individual experts.
Start from the customer outcome and reason backward to today.
Limited capacity holding-tank for current information.
We prefer eliminating a small risk completely over reducing a larger one partially.
One party's gain is exactly another party's loss.