| Ruby | v1.6.0Sep 16th 2020 | (+14) | ||
| Zack_M_Davis | v1.5.0Nov 17th 2009 | (-21) byline removal | ||
| Vladimir_Nesov | v1.4.0Oct 27th 2009 | (+5/-25) double newlines look ugly | ||
| Zack_M_Davis | v1.3.0Oct 27th 2009 | (+93) locate the hyp., Einstein's arrogance | ||
| PeerInfinity | v1.2.0Oct 21st 2009 | (+10) | ||
| Zack_M_Davis | v1.1.0Oct 21st 2009 | (+695) wrote page | ||
| Eliezer Yudkowsky | v1.0.0Sep 28th 2009 | (+40) created stub |
To a Bayesian, evidence is a quantitative concept. In principle, a hypothetical Bayesian superintelligence could say not only that the evidence supports a particular hypothesis, but by how much. (In practice the true math is usually intractable.) The more complicated or a priori improbable a hypothesis is, the more evidence you need just to justify it, or even just single it out of the amongst the mass of competing theories.
We often find it convenient to express the amount of evidence in terms of logarithms of odds, decibels if we use the base-10 logarithm, bits if we use the base-2.
A Technical Explanation of Technical Explanation by Eliezer Yudkowsky