GM Andrew Tang Defends Humanity Against Leela Chess Zero

Hey @arex, great write-up of a fantastic event! I think one thing I would add is that only in the games without increment, Andrew lost many games on time that he may have won otherwise. For example, there was another ultrabullet game where Leela blundered a queen to the same Bxh7 sacrifice, but where Andrew lacked the time to convert.

This point is a bit irrelevant to the discussion of Leela's thinking speed, but the fact that she can move instantaneously definitely gave her the edge over Andrew, since she converted many endgames by timing out the human in positions he knew were drawn.

Cool writeup, I enjoyed it. It's a shame these game boards don't have real time replay speeds though! GG Leela

I think the article should mention that ID 125 wasn't even the strongest at the time, if only to give readers a feel for Leela's continuous progress.

You both make good points, @FAUM and @Veletas

4:44 is defending humanity? It's Leela today and T-800 tomorrow.

As algorithms become more complex and the static nature of the (although numerous) but ultimately limited variables involved in Chess (the moves) then when you add in risk vs reward AI training and ever increasing computational power, it's only a matter of time until there is a Chess program that is unbeatable even by the GM's or someone rocking a FIDE rating of 3000+

дааа...

@arex when someone tells 'A defends B against C', it usually means something like 'C's attempt to do something against B was unsuccessful thanks to A'.

Judging by the score: 1 win, 6 draws and 37 losses, it looks like the correct title should be something along the lines: "A FAILS to defend B against C".

Or was the goal of LC0 was to win all matches and AT made this impossible by a win and a few draws?

LC0 is a new opponent and we don't know much about her. As such it's unfair to judge all humanity by the result of a 1-hour match.

First thing, I find it strange that people suddenly started referring to *computer programs* as "he"/"she". It's just a sequence of bytes after all.

Second: did I get it right that it had not surpassed Stockfish yet?
And doesn't Stockfish also use genetic programming approach to tune numeric parameters used in material and position evaluation, etc.? (thereby also being self-taught to some extent)