A few books at various levels with very high signal-to-noise ratio: Feynman's little popular science book _QED: The Strange Theory of Light & Matter_ is a gem. It's very short and non-mathematical, but still manages to get remarkably close to the truth. If you're more mathematically inclined (e.g., fluent in multivariate calculus) and know a little classical physics, the first 5 chapters of Wald's _General Relativity_ are an essentially complete introduction to the subject. Another classic is Shannon's _A Mathematical Theory of Communication_, which introduces the subject of information theory, and then solves basically all the subject's main problems. In economics, I think the record for signal to noise is probably held by Keynes _The Econonic Consequences of the Peace_, 118 pages, in which he basically forecasts the disastrous course of events from 1918 to 1939.
Also recommended in this reply