A Fervent defense of Frequentist Statistics. Eliezer Yudkowsky's comments are also well worth reading. Most notably:

Who are these mysterious straw Bayesians who refuse to use algorithms that work well and could easily turn out to have a good explanation later? Bayes is epistemological background not a toolbox of algorithms.

After thinking about the online learning example discussed in the above post, I came to the realization that Eliezer Yudkowsky came to a long time back. Randomization (aka "non-bayesian") algorithms are effective in adversarial problems not because Bayesian reasonining fails, but because randomization reduces the advantage that your adversaries intelligence provides. The adversarial bandit is fundamentally not a statistics problem at all.

Scala's Types of Types. Great article explaining Scala's type system.

Peter Thiel is wrong about the minimum wage. This example just goes to show why a concrete model is so important - Peter Thiel is a very smart guy, yet his verbal reasoning is easily debunked by a very simple graph.

Hip Gadgets For The Developing World Won't Solve Global Poverty. See also this article.

Republicans understand evolution better than democrats. They don't *agree* with it, but they understand it. Republicans are also more likely to know that the Earth revolves around the Sun once every year. Stereotype busted, I guess.

DO NOT USE CONFIDENCE INTERVALS. See the paper Robust Misinterpretation of Confidence Intervals. They are a *useless* tool for communicating with non-statisticians - anyone without a PhD will interpret your confidence interval as a credible interval. I've all observed this in practice, but it's good to have stats to back it up.