Definition
Plain language
Your updated belief about something after you've taken new evidence into account.
As stated in the literature
In Bayesian inference, the distribution over a quantity after conditioning on observed evidence, proportional to prior times likelihood; the RL-as-inference view identifies the optimal KL-regularized policy with a reward-tilted posterior over the base model.
Why it matters: The posterior is the payoff of evidence-based reasoning — the belief you should act on once the data is in.
For example, after a medical test comes back positive, your posterior is the updated chance you actually have the condition, combining the test result with how common it is.
Heard on the show
“Take the answer, with majority voting over a few samples for noise robustness, and update the posterior over the candidates using Bayes' rule.”Episode 091 — When Better Fine-Tuning Can't Help: A Geometric Impossibility in LLM Causal Reasoning