Sophros
How to Learn Complex Topics

Bayesian Thinking for Beginners: Update, Do Not Reset

Learn priors, likelihoods, posteriors, base rates, and why evidence should update confidence rather than flip beliefs instantly.

Published September 30, 20264 min read

Evidence does not arrive in a vacuum. Its meaning depends on what was plausible before it appeared.

Quick verdict: Bayesian thinking combines prior plausibility with how expected the evidence is under competing explanations. Strong evidence can move belief sharply; weak evidence should not.

Begin with base rates

If a condition is rare, even a good test can produce many false positives relative to true positives. Prevalence belongs in the interpretation.

Compare likelihoods

Ask how probable the observation would be if each hypothesis were true. Evidence favors the hypothesis that makes the observation less surprising.

Posteriors become new priors

Updating is sequential. A conclusion remains provisional because tomorrow’s evidence begins from today’s revised confidence.

A medical test without algebra

Among 1,000 people, suppose 10 have a condition. A 90%-sensitive test finds 9; a 5% false-positive rate flags about 50 others. A positive result means roughly 9 of 59 positives are true—not 90%.

What to remember

  • Base rates shape evidence.
  • Likelihood asks which explanation predicts the observation.
  • Beliefs should move by degrees.

Test whether you understood it

A colleague succeeds after adopting a new routine. List the prior success rate, alternative causes, and evidence that would distinguish the routine’s effect from coincidence.

Where Sophros fits

Sophros can sequence intuition before notation, then connect Bayes to diagnosis, forecasting, science, and everyday decisions in 3–15 minute lessons.

Sophros.me builds connected, narrative-driven courses around the question you choose. Each lesson can be set from 3 to 15 minutes, which makes the format useful for a commute, a break, or a deliberate return to reading. Generated material can contain errors, so consequential claims should be checked against primary or authoritative sources.

A practical next step

  1. State competing hypotheses.
  2. Estimate the base rate.
  3. Ask which hypothesis predicts the evidence better.

The goal is not to collect one more finished page. It is to leave with a model you can explain without the page open. Close the tab, write the central idea in your own words, and name the question that remains unresolved. That small act separates learning from smooth consumption.

Sources

Straight answers

Frequently asked questions

Try the different model

Turn this question into a course you can continue tomorrow.

Choose the exact angle and reading time. Sophros builds a connected 3–15 minute learning path around it.

Build a learning path
Keep pulling the thread
Browse all guides