Learn it → Apply it → Save as Skill
There is a ceiling on how many people you can actually, mutually know. Past roughly 150 names, the bookkeeping in your head runs out: you can still recognize a face or recall a title, but you can no longer track who that person is to everyone else around them. Headcount keeps climbing. Real, maintained relationships do not.
The Dr. Wallace’s Favorite Mental Models is designed to turn useful ways of thinking into reusable tools. Each model gives you a fast visual overview, deeper interactive learning, practical application exercises and an AI-ready Markdown skill you can save for future use.
Instead of merely reading about mental models, you can gradually build your own AI skills library: a collection of portable reasoning skills that can help ChatGPT, Claude, Grok and other AI assistants apply the mental models you value.
Anthropologist Robin Dunbar found that primate species with larger neocortices tend to live in larger stable social groups, and extrapolated the human ratio to a predicted group size of about 150: the number of people whose identity and relationship to everyone else you can hold in mind at once.
Important: each layer is an approximate order of magnitude, not a precise headcount. Treat 150 as the most commonly cited layer, not the only one that matters.
The model is most useful whenever headcount, membership, or a contact list has grown past the point where everyone can still track who everyone else is, without anyone having decided that on purpose.
The method separates how many people a structure currently asks someone to track from how many a person can actually track at that layer.
A team, community, or contact list has crossed roughly 150 (or another layer boundary) and coordination is starting to fray.
Sort the people involved into the nested circles: about 5 intimate, 15 close, 50 trusted, 150 stable network, 500 acquaintances, 1500 recognizable.
Compare the structure you are running, one flat team, one shared channel, one account book, against the layer it is actually asking people to maintain informally.
Split, tier, or add deliberate structure, sub-groups, moderators, account managers, so no one is required to hold more relationships than that layer allows unaided.
Open a use case to see what the model changes.
As headcount climbs from a startup-sized team toward a few hundred, people stop recognizing colleagues in the hallway and start needing org charts to know who does what. One manufacturer has long capped individual plant headcount near 150 and built a new plant rather than expand one further, choosing to replicate rather than enlarge.
A community that felt close-knit at 80 members can feel like a crowd of strangers at 2,000. Deliberately built sub-channels, cohorts, or moderator-led groups can recreate a layer small enough for members to actually know each other again.
A CRM can store unlimited contacts, but the depth of relationship a single person can maintain is still bounded. Tiering accounts, high-touch, mid-touch, self-serve, matches coverage to the layer a human can realistically sustain rather than assuming software removes the limit.
Historical military units, such as the Roman century, were sized well below Dunbar’s number, letting a commander know every subordinate personally. Modern units nest smaller groups inside larger ones for the same reason: command and trust work best at a size a person can actually track.
A phone with 2,000 names is not 2,000 relationships. Separating the inner layers, who you would call at 3am, your close circle, your regular contacts, from the long tail of names you merely recognize, clarifies where your limited relationship-maintenance time actually goes.
Designing a “close friends” feature, a team channel cap, or a recommended group size can draw on the nested layers directly, rather than guessing at an arbitrary number or assuming larger is always better engagement.
This is an illustrative example designed to teach the method.
Expand for where the figure comes from and how confidently it has held up.
British anthropologist and evolutionary psychologist Robin Dunbar published the original correlation in a 1992 paper in the Journal of Human Evolution, relating the ratio of neocortex volume to total brain size across primate species to each species’ typical social group size. Extrapolating the human neocortex ratio along that relationship predicted a group size of roughly 150.
No. Dunbar himself presented it as an approximate figure, and a 2021 statistical reanalysis of the original primate dataset, led by researchers including Patrik Lindenfors, argued that the method supports a very wide confidence interval, from roughly the dozens up into the low thousands. That means 150 is a reasonable point estimate inside a wide range, not a tightly measured number.
Dunbar and later writers pointed to converging real-world figures offered as circumstantial support: average sizes of hunter-gatherer bands and historical village communities, Hutterite farming settlements that have long tended to split once they approach 150 members, and the size of some historical military units. None of these prove the figure on their own, but they are commonly cited alongside it.
The underlying claim is an approximate, layered range. Nothing distinct happens the instant a group reaches person number 151.
A list of 2,000 names is not 2,000 stable relationships. Most of those names sit in the outer, shallower layers, recognized but not actively maintained.
Fixating only on the headline 150 misses that the 5, 15 and 50 layers, your closest circle, your close friends, your trusted group, usually matter more day to day.
Social platforms can expand how many faces or names you recognize. Researchers have found far less evidence that they expand how many relationships people actively, mutually maintain.
The model explains why deliberate structure becomes necessary past a given size. It is not an argument for staying permanently small when growth genuinely serves the goal.
A small team doing novel, high-trust work tolerates far less headcount growth before fraying than a larger unit doing routine, well-documented, loosely coupled work.
the goal is information flow, job leads, or reach, not relationship depth. Large, shallow networks can spread information efficiently precisely because they are not trying to be 150 close friends.
the organization already runs on explicit roles, documentation, and systems, a hospital, an airline, rather than depending on everyone personally knowing everyone.
the relationship is one-to-many by design, an author and readers, a creator and followers, rather than a claimed mutual, two-way social bond.
This browser tool structures your thinking. It does not send your information anywhere or perform AI analysis.
Save as Skill converts the model into portable Markdown instructions that teach an AI when and how to apply Dunbar’s Number, what questions to ask, what evidence to seek, what mistakes to avoid and how to structure its analysis.
Save the skills you find useful and organize them into your own reusable library. The Markdown is intentionally model-neutral, so it can be used wherever an AI system accepts uploaded files, project knowledge, custom instructions or reusable text context. Platform capabilities differ, so this is a portable skill file rather than a claim of one-click installation.
No. It describes the number of relationships you can actively, mutually maintain, not the number of names a platform lets you add. Research has not reliably found that a larger online friend list reflects a larger number of real, sustained relationships.
Dunbar and colleagues observed that each wider circle of relationship tends to be a few times larger than the one inside it, forming a nested structure rather than one flat number. The 150 figure is the most commonly cited layer, but the inner layers often matter more for day-to-day decisions.
It remains debated. Some researchers find the historical and organizational evidence broadly consistent with a number in that range. A 2021 statistical reanalysis of Dunbar’s original primate dataset argued the method could support figures anywhere from well under 100 to several hundred, so the number is best treated as an order-of-magnitude heuristic rather than a precise measurement.