priors
The initial probability assignments to hypotheses before new evidence is incorporated in Bayesian reasoning.
As defined in ms_ama_february_2024
Where it comes up · 19
- Noah Smith on Whether Economics is a Science4 claims
- AMA | April 20224 claims
- Nate Silver3 claims
- 329 | Steven Pinker on Rationality and Common Knowledge3 claims
- Demis Hassabis — Scaling, superhuman AIs, AlphaZero atop LLMs, AlphaFold2 claims
- Rob Wiblin and Russ Roberts on Charity, Science, and Utilitarianism2 claims
- David McRaney on How Minds Change2 claims
- 478 - The Psychedelic Mind2 claims
and 11 more
Most important claims · 10 of 34Search all claims for “priors” →
- Naive realism is the assumption that all people need to do is see what you have seen and they will naturally agree with you; it reflects a lack of cognitive empathy and ignores that other people come from completely different priors, experiences, and social influences that shape both how they form beliefs and how they interpret evidence.David McRaney on How Minds Change
- Finding life on Europa or Enceladus would substantially raise credence that life is frequent because the likelihood of such a finding is far higher in a life-rich universe than a life-poor one, but you cannot draw the conclusion from Bayesian reasoning alone—you must state your priors, since with a sufficiently tiny prior the update would be negligible; conclusions depend on both likelihood functions and priors.AMA | April 2022
- The questioner asks whether the host assigns explicit probabilities and updates them via Bayes' theorem, noting that a 70 percent forecast is judged wrong when the event fails, which makes percentage framing risky.Mindscape Ask Me Anything, Sean Carroll | September 2026
- Children are actually better than adults at reaching unlikely new solutions because, in Bayesian terms, adults' priors become more peaked and confident, making it harder to conceive of new possibilities; childhood itself can be understood as a solution to the explore-exploit tension, dedicating a protected life-history phase to wide exploration before narrowing in.153 - Possible Minds
- When a prediction fails, a sufficiently strong prior lets a modeler add a new assumption or tweak the model to explain away the contradiction and preserve cherished assumptions; doing this every time a large natural experiment fails violates scientific method and means the view is unfalsifiable.Paul Pfleiderer on the Misuse of Economic Models
- You cannot replace choosing a prior with another Bayesian sub-problem pushed up a stack to a foundational edge, because priors are starting points reflecting your current credences, not finishing points; you can argue for them only by relating one prior to another and checking internal coherence, but ultimately humans are embedded in the world with pictures of what makes sense, and grounding priors in some absolute foundation is the wrong goal—it is a feature, not a bug.AMA | April 2022
- Ideology in economics is essentially wishful thinking—people derive their priors from what they want to be true—and because we don't always know which policies work or which evidence is right, economists must rely heavily on these priors, so bias inevitably plays a role in policy debates.Noah Smith on Whether Economics is a Science
- When a strong positive attitude (toward a political candidate) collides with a strongly negative event, people resolve the dissonance in ways that feel like NOT changing their mind—reinterpreting the bad act as actually positive, denying it happened, or claiming the actors weren't really their side—yet they have in fact updated their priors and beliefs about the situation.David McRaney on How Minds Change
- While a pure-RL, no-priors AlphaZero-style path to AGI is theoretically possible, the quickest and most plausible path is to use all existing world knowledge from the Web via scalable algorithms like transformers as a prior, then add planning and search on top; the final AGI will likely include large multimodal models but not rely on them alone.Demis Hassabis — Scaling, superhuman AIs, AlphaZero atop LLMs, AlphaFold
- Natural experiments are extremely hard to find in macroeconomics because macro phenomena are global rather than bounded—there are no clean borders where a policy works in some places and not others, and even cross-country comparisons are contaminated by linkages—so macro is almost by definition the domain where data is worst, which is precisely why priors and ideology play a larger role in macro debates.Noah Smith on Whether Economics is a Science