317 karma · joined November 11, 2016
Technical Director @ Team Liquid. Live in Berlin, sometimes around Utrecht! I'm interested in statistics in competitive video gaming, mostly Dota 2; as well as mathematics, film, and trivia. I run datdota.com and windrun.io
The prior vote was relatively close (55/45) so it's reasonable to expect that 200k people might've flipped to the 'Yes' camp.
If you filter on only valid area codes, the number is a bit smaller (620350), and if you exclude Toll Free, Reserved, and Government area codes it's down to 567277.
I thought that they were suggesting it's ~ the average winrate given optimal play.
> and otherwise fully detail their "we throw out anyone that wins" policy up-front
They can just ask the gambler to leave the premises - I cannot imagine some 'libertarian free market' solution which would force casinos to allow anyone to enter the premises and gamble (provided it's not clear discrimination). Being able to count cards isn't some sort of protected class, right?
They have two lists of words: one is a list of possible answers, and another is a list of extra (valid) words which can be guessed (in addition to the words in the answer list). Sometimes it might be better to use a non-answer word as a guess: the best case gets worse (since you cannot win immediately) but the average case and worst case both get better.
This isn't strictly true either. Two N-sized subsets of words from a common initial set might have completely different difficulty in reducing further, because in the worst case there might not be a valid guess which nicely spreads the remaining words out among the 243 possible outcomes for that guess.
Set-size is a good heuristic, but it's just that - a heuristic.
I also just rsync my music library to an external SSD so that it's backed up, and up-to-date when I'm travelling.
Also interested why they went with Trueskill over Glicko 2, since it's just 1v1 'encounters' anyway.
I'm also not a huge Maurice Ashley fan - but think he did a good job handling the press conferences. What was interesting was that for Karjakin/Carlsen, Jan Gustafsson was doing commentary, but also was part of the Carlsen analytics/prep/second team. Svidler's analysis is always enjoyable, but sometimes he seems bored with games and his language almost becomes 'this is over let's move on'.
Dropping Elo after winning an event isn't really that 'wonky' - it's just a Bayesian update for underperforming your statistical expectations. The same thing happened when Carlsen won the Candidates in 2013, he went +5=7-2, but he was 2872 rating against a field of 2774, so 8.5/14 would mean he loses ~6 rating.
Non-NN chess engines are really all about doing a good job at balancing these heuristics (like it says above). There's fundamentally an upper bound in how good the overall approach can be, simply because the underlying heuristics are defined and selected by humans: there could be many more heuristics that haven't been considered. This version of "learning" is really just removing human forced constraints which were introduced to help learning, by starting from a position without bias.