En 30 segundos
A worked example, a probability formula and a square diagram explain how Bayes’ theorem combines prior probability with evidence. The video also discusses ambiguity in the assumptions behind the example.
- 1Count the matching casesFour of 24 people matching the illustrative description are librarians.
- 2Combine prior and likelihoodMultiply the prior by the likelihood, then normalize by the total evidence probability.
- 3Use area as probabilityA square diagram turns the calculation into a ratio of regions.
- 4Inspect the assumptionsSelection and evidence generation determine the appropriate inputs.
Conclusiones
- Evidence updates a prior; base rates matter.
- Posterior probability is the share of evidence cases satisfying the hypothesis.
- Areas in a square can visualize the calculation.
- The numerical result depends on assumptions about selection and evidence.
Lee el texto editado por capítulos
Steve and base rates
Bayes’ theorem describes how evidence updates a prior probability. The video aims to build an intuitive and geometric understanding before presenting the formula.
Bayes’ theorem describes how evidence updates a prior probability. The video aims to build an intuitive and geometric understanding before presenting the formula.
The speaker introduces Steve, described as shy, orderly and attentive to detail, and asks whether he is more likely to be a librarian or a farmer. The example comes from work by Kahneman and Tversky.
A stereotype might favor librarian, but the base rates also matter. The example uses an assumed ratio of 20 farmers for every librarian; it is an illustration, not a current labor statistic.
Imagine 200 farmers and 10 librarians. If the description fits 10% of farmers and 40% of librarians, it fits 20 farmers and four librarians; among these 24 people, four are librarians.
Under those assumptions the probability is 4/24, about 16.7%. Even a description four times as likely for librarians does not overcome the assumed 20-to-1 base rate.
The key idea is that evidence updates a prior belief rather than determining the answer in isolation. Different reasonable inputs can yield different numerical answers.
Bayes’ formula
Let H be a hypothesis and E the observed evidence. The conditional probability P(H|E) asks how likely H is after restricting attention to cases where E holds.
Let H be a hypothesis and E the observed evidence. The conditional probability P(H|E) asks how likely H is after restricting attention to cases where E holds.
The numerator P(E|H)P(H) is the probability of both the hypothesis and the evidence. The denominator P(E) includes evidence arising both when H holds and when it does not.
Bayes’ formula is P(H|E) = P(E|H)P(H)/P(E), provided P(E) is positive. It expresses the fraction of evidence cases in which the hypothesis is also true.
The speaker notes uses in science, machine learning and explicit probabilistic models. The same arithmetic applies whenever a prior and a likelihood are combined to update a belief.
A geometric view
A unit square provides a visual: its width represents the prior probability of H, and the height within each region represents the probability of evidence under that condition.
A unit square provides a visual: its width represents the prior probability of H, and the height within each region represents the probability of evidence under that condition.
The area of the overlap is P(H)P(E|H). Comparing it with the total area where evidence holds reconstructs the posterior probability.
Intuition and conjunction
A second example, the Linda problem, illustrates the conjunction fallacy: a person cannot be more likely to be both a bank teller and a feminist than to be a bank teller.
A second example, the Linda problem, illustrates the conjunction fallacy: a person cannot be more likely to be both a bank teller and a feminist than to be a bank teller.
Presenting the same question as counts out of 100 can make the inclusion relationship easier to see. The video uses frequency framing to connect intuition to the probability rule.
The square and the counting approach both restrict attention to cases with the evidence, then ask what fraction also satisfies the hypothesis. They are two views of the same calculation.
Ambiguity in the example
The Steve example is open to debate because the way Steve was selected and the meaning of the description affect the appropriate prior and likelihood. Bayes’ rule does not supply those assumptions by itself.
The Steve example is open to debate because the way Steve was selected and the meaning of the description affect the appropriate prior and likelihood. Bayes’ rule does not supply those assumptions by itself.
Más de este canal
40:44EN
3Blue1Brown· 16 oct 2026Grant Sanderson: There's never been a better time to go into mathGrant Sanderson, Constantin Kogler · 3Blue1BrownStudents fear AI has solved mathematics. Grant Sanderson argues the opposite: even if machines prove theorems better than we do, the institution of math stays valuable—and this is a rare moment when newcomers can reshape it.
51:55EN3Blue1Brown· 18 sept 2026The last IMO problem AI could not solveGrant Sanderson · 3Blue1BrownGrant Sanderson walks through IMO 2025 Problem 6 — a tiling puzzle that frontier models missed — and the visual patience it takes to prove a construction is optimal.
0:55EN3Blue1Brown· 24 jul 2026The 64 Sugar Cubes Puzzle: Six Equal DistancesA 4×4×4 block has three colors; prove that six same-colored pairs share a distance.
33:51EN3Blue1Brown· 16 jul 2026What Cross-Entropy Connects: Compression, Language Models and KL DivergenceGrant Sanderson · 3Blue1BrownGrant Sanderson builds cross-entropy from code lengths and uses it to explain next-token training and distillation.
2:25EN3Blue1Brown· 12 jun 2026How Shannon Measured the Entropy of EnglishA guessing game helped Claude Shannon estimate how much information English text contains, and why prediction is central to compression.
1:02EN3Blue1Brown· 10 jun 2026A Robot Walk and the Puzzle of Optimal Binary EncodingHow few bits does it take to send directions that occur with unequal frequencies?