That's a good summary. Their description of Petrosian is laughable and is also contrary to Schaeffer's.

In this game it seems that Stockfish had played Nd6-b7 planning d5, unwinding ASAP, but was never given the chance to play it because it missed Bg5: hence the terrible lack of development of the queenside.

So while Bg5 will get most of the plaudits, it's the pawn sacrifices for piece activity that are particularly impressive and dare I say it, human-like (without wishing to insult AZ).

Stockfish wasn't being run under ideal conditions, and a video I saw from an IM who does a lot of deep analysis with Stockfish claimed that Stockfish would have drawn 7 or 8 of the 10 sample games under TCEC conditions. The full Bg5 game is here, this IM claimed that Stockfish's losing move was 17. ... Qg6, which leaves the queen too vulnerable, and that instead Qd8 (preferred by Stockfish with deeper search) would hold the position. I have no idea if that's true or not.

It's not really the point whether AlphaZero is stronger than Stockfish or not though, the points are:

- Stockfish was designed by humans to play chess, AlphaZero is a general purpose neural network that can learn to play many different games
- Stockfish has been honed over tens of thousands of hours of self-play and tens of thousands more hours of programmer work, AlphaZero learnt in 4 hours (albeit with an enormous amount of computing power)
- When you look at AlphaZero's games, there's clear direction to its moves. It's evident that positional understanding is built into it. Games from other engines can often look directionless, with the engines shuffling pieces around and having no idea how to proceed. When AlphaZero drops one of its horrifying positional binds, you can look back in the game and see it patiently constructing the position well in advance. My favourite game is this one, where Stockfish ends up so locked up that it has to give up its queen to get out of the bind. There, the fantastic move 37. Bd1! reroutes the bishop to the b2-g8 diagonal well in advance of any appearance of usefulness there. Fast forward 10 moves to move 47 and suddenly the bishop is the star performer in the position.

What would go wrong if Stockfish took the bishop with its pawn?

I've watched all of Jerry's AZ vs SF youtubes (Chess Network), and I'm agog at AZ's play after just a few hours of self-training.

The question still remains (as I posted in the other thread) of why the other 90 games haven't been released. I'm guessing that if it had been given 4000 hours for self-training instead of 4 hours we'd have seen more of these, so I'm hoping they do this because I'd love to see how it treats certain openings (is the Chigorin Spanish good or bad for black?).

Much stronger players than I are saying that after the recapture with the knight, Be4 with an attack on the queen (on d3 or c2) and on h7 looks fatal (compare attackers with defenders) - but apparently people's desktop Stockfish isn't seeing the seriousness of this until several moves deeper than its evaluation in the actual game (the horizon effect).

f5 and d5 are supposed to offer sterner resistance than the bishop capture, but prove futile after a further 20+ ply or so, according to some heavyweight computer analysis.

But we're used to computers flashing out very long winning combinations. What's unique here for a computer is the scant regard that AZ has for material.

Speed estimates? I'm going with mid-80's.

Street view

aliens

Latest Space X fail

Santa Ana winds out of control this year.

I'd take the over on 100mph.

Schaeffer also draws a comparison with Tigran Petrosian, a Soviet champion from the 1960s who played an attack-on-all-fronts game: “He was like a python — he would slowly squeeze you.”

Double fail; Pertrosian did play a patient 'pythonesque' game. They must have confused Tigran with Tal, the 60's Soviet champ who successfully played the all-out attacking game.

Nxg5 forces Qg8, because on any other queen move, Be4 evicts the queen from the diagonal and leaves Black's king totally defenceless. (Edit: Or Qh6, Nxf7+ where Rxf7 gets Black mated after Re8+). So after Qg8, White plays Qh4.



It gets complicated from there but you can see that Black is in huge trouble. All his pieces are jammed in each other's way, h6 is coming (where if g6, White switches the queen onto the a1-h8 diagonal and mates Black on g7) and so are moves like Re7 and Be4, with White's pieces all descending on Black's king. As an example, say f6 to evict the knight. There's no time for this. h6 anyway and if fxg5, hxg7 double check, Black's king is exposed, it's a forced mate in 6 from there.

Stockfish sees ~immediately that hxg5 is not tenable. The problem with f5 is buried much deeper down and is the reason it takes so long to find Bg5.

I’d guess 3 grams

Nice, that's nice.

Thanks jalfrezi and Chris for the explanations.

Stay classy bro

when I look at the position in post 913 my first thought is how many moves can I take to get a knight to c2 to pin the rooms.

It kind of sucks to know your families contribution to chess was defeated by a computer using brute calculations and tress that has now been defeated by a nueral algorithm that no one can predict how it thinks and then a bunch of uncreative peons use computer assisted analysis to sound smart about something they'll never really understand not that it bothers me anymore. **** computers though.

It's like my dad told me once . It's good advice and one of those things I understand that second class posters like el diablo and oot clique don't really but they might some day: if you live long enough no matter what you know , no matter how valuable that knowledge is , one day no one will care. On top of that I'm insane and banished and working at a grocery store in Kansas . Literally no one could no this situation. Not literally no one. Hopefully if I drink enough vodka over the next five years I can eliminate myself from the list too.

I assume it beats working at a Cinnabon in Nebraska.

I'd probably have an excellent job at a Cinnabon in Nebraska.

That‘s a fork, not a pin.

By the way, there are big diminishing returns on training for neural networks, so training for 4000 hours probably wouldn't improve it much. I'm not an expert or anything but I think the way to make it stronger would be to throw more hardware at it. A0 trained on 64 second-gen TPUs but when it played, it only ran on 4. There are some problems with running game tree searches in parallel, but throwing more processors at it would definitely help.

Everyone Is Getting Hilariously Rich and You’re Not

While this is specifically about bitcoin millionaires it feels like this title really captures the world at large right now