prior
In Bayesian inference, the initial probability distribution over hypotheses before observing evidence; must be broad enough to include all possibilities.
As defined in ms_94
Where it comes up · 15
- 320 - Constructing Self And World7 claims
- 143 | Julia Galef on Openness, Bias, and Rationality4 claims
- 335 | Andrew Jaffe on Models, Probability, and the Universe4 claims
- AMA | August 20214 claims
- Some thoughts on the Sutton interview3 claims
- AMA | March 20223 claims
- 6 | Liv Boeree on Poker, Aliens, and Thinking in Probabilities2 claims
- AMA | June 20212 claims
and 7 more
Most important claims · 10 of 36Search all claims for “prior” →
- The slogan 'follow the science' is unscientific because it obscures the question 'which scientist?'—scientific consensus is a reasonable Bayesian prior but not infallible, and treating dissent as prima facie evidence of lunacy contradicts the scientific method; the proper response to an open question is a careful evidence-weighing deep dive, not demanding that society accept claims on faith because of scientific credentials.Megan McArdle on the Oedipus Trap
- Bayesian analysis is updating your model of the world based on new information, shifting your beliefs by an amount proportional to the strength of the evidence—as when seeing a tree with purple leaves shifts you away from the belief that all leaves are green—and Bayes' theorem provides the mathematically correct, rational way to do this updating.6 | Liv Boeree on Poker, Aliens, and Thinking in Probabilities
- Bayes' theorem is a mathematical theorem you cannot disagree with given the axioms, but Bayesian epistemology is a broader claim that updating credences from priors is how we should gain knowledge; new insights, not just new empirical data, count as new data that can change your credences.AMA | March 2022
- One should never set a Bayesian prior to 1 (or 0), because a prior of 1 means no amount of evidence can ever change your credence; the special problem with prior 1 is distinct from merely updating a non-extreme prior toward zero given evidence.AMA | September 2024
- Everyone is entitled to their own priors but no one is entitled to their own likelihoods; priors come from human brains embedded in the physical world that simply have opinions, and the hope of Bayesian epistemology is that enough data collection washes out differing priors.AMA | March 2022
- Culture exerts an overwhelming, largely invisible constraint on experience by setting high-level priors: the same perception (e.g. an exposed or covered face) is shocking or unremarkable depending entirely on cultural expectation, and these encodings are taken for granted like water to a fish.320 - Constructing Self And World
- Whether a significant small-sample result is a complete false positive or merely an exaggerated true effect depends on the field's prior probability of real effects: in fields with strong theory and prior evidence you tend to get inflated-but-real effects, whereas in agnostic, high-noise fields searching in the dark, almost any significant result is a false positive.John Ioannidis on Statistical Significance, Economics, and Replication
- A single measurement can yield negative 'specific information' even though the average information gained from a set of measurements is positive: for example, not finding your keys in your most-likely pocket leaves you less certain of their location than before, so that specific outcome carries negative information.266 | Christoph Adami on How Information Makes Sense of Biology
- The problem of induction has no deductive solution, but Bayesian inference is its answer: you start from meager prior information (itself a model), and systematically update with data to become more and more certain, though you can never reach 100% certainty.335 | Andrew Jaffe on Models, Probability, and the Universe
- A Bayesian approach to knowledge increases confidence that one is approaching truth without ever reaching certain truth statements, but Popperian epistemology argues Bayesian processes ultimately rest on an initial guess or conjecture forming the prior, so conjecture and refutation may be the more fundamental truth-seeking process—or alternatively, Bayesian updating could itself be construed as part of the critical testing demanded by conjecture and refutation.AMA | August 2025