Re: adversarial ML & steganography. There were 2 papers at this year's NIPS studying these ideas (http://papers.nips.cc/paper/6802-hiding-images-in-plain-sigh... and http://papers.nips.cc/paper/6791-generating-steganographic-i...)

This is also relevant to your comment: https://arxiv.org/abs/1705.07663

4RR2/1p4p1/p1p5/2p3p1/2P2kP1/7P/PP1r1P1K/8 b - - 2 31

Cool project but not hard to beat.

Very curious as to why this so much more responsive than overleaf and sharelatex. Anyone?

I'd guess that the rendering is done client side here and can be done with partial updates. ShareLaTeX and Overleaf have to do a full LaTeX pass on the server just to show the change of a small part of the document.

Yeah. Browse the web with multiple tabs and every current fingerprinting attack is useless.

Really? I'd argue 1% of my income making minimum wage is far less disposable and than a 1% of the one percents salary. And so I would be far less inclined to buy more votes. Then taking your Euro analog, is it right/fair to make someone pay more (even say in proportion to their salary) than any other person for a vote? Who decides how much I should pay for my vote? Is it based on current salary? Because if it is there are a whole load of very rich people who technically have a very small salary. Is it based on net worth? Sounds like a nightmare to me.

Approximate solutions are better than clearly wrong solutions.

Nothing really trumps this article on big numbers. http://www.scottaaronson.com/writings/bignumbers.html

Won't these high capacity middle relays eventually become fast guards (and because of the new flag assignment process, possibly both guards and exit relays simultaneously) or can the directory authorities restrict flag assignments even if the relay is eligible? By "can" I mean "should they" I suppose?