Spectator chess pieces met with Sci-Fi Pawn Play! Komodo 9.02 vs Leela ID 511

♚COURSES kingscrusher.t...
📚 kingscrusher.t...
📚 kingscrusher.t... || kingscrusher.t...
#KCLeelaChess
♚ Play turn style chess at bit.ly/chessworld
🎥 Immortal Must See Chess Games: bit.ly/2KiWGhP
🎥 Leela Chess Playlist: bit.ly/leelachess
👕 Kingscrusher Store :teespring.com/...
♚Support the channel by donating via Paypal: goo.gl/7HJcDq
🎥 Immortal Must See Chess Games: bit.ly/2KiWGhP
Game quality tags: amazing, awesome, astonishing, brilliant, classic, crushing, dynamic, elegant, exceptional, excellent, exciting, fabulous, famous, fantastic, finest, flashy, greatest, important, impressive, incredible, instructive, incredible, interesting, magnificent, marvellous.
Info about Leela Zero:
en.wikipedia.o...
...
Leela Chess Zero (lc0) is a free, open-source, and neural network-based chess engine and distributed computing project.
Leela Zero's algorithm is based on DeepMind's 2017 paper about AlphaGo Zero.[3][6] Unlike the original Leela, which has a lot of human knowledge and heuristics programmed into it, Leela Zero only knows the basic rules and nothing more.[7]
Leela Zero is trained by a distributed effort, which is coordinated at the Leela Zero website. Members of the community provide computing resources by running the client, which generates self-play games and submits them to the server. The self-play games are used to train newer networks. Generally, over 500 clients have connected to the server to contribute resources.[7] The community has provided high quality code contributions as well.[7]
Leela Zero finished third at the BerryGenomics Cup World AI Go Tournament in Fuzhou, Fujian, China on 28 April 2018.[8]
Info about Alphazero:
en.wikipedia.o...
AlphaZero is a computer program developed by the Alphabet-owned AI research company DeepMind, which uses an approach similar to AlphaGo Zero's to master not just Go, but also chess and shogi. On December 5, 2017 the DeepMind team released a preprint introducing AlphaZero, which, within 24 hours, achieved a superhuman level of play in these three games by defeating world-champion programs, Stockfish, elmo, and the 3-day version of AlphaGo Zero, in each case making use of custom tensor processing units (TPUs) that the Google programs were optimized to make use of.[1] AlphaZero was trained solely via "self-play" using 5,000 first-generation TPUs to generate the games and 64 second-generation TPUs to train the neural networks, all in parallel, with no access to opening books or endgame tables. After just four hours of training, DeepMind estimated AlphaZero was playing at a higher Elo rating than Stockfish; after 9 hours of training, the algorithm decisively defeated Stockfish 8 in a time-controlled 100-game tournament (28 wins, 0 losses, and 72 draws).[1][2][3] The trained algorithm played on a single machine with four TPUs.
...
Relation to AlphaGo Zero
Further information: AlphaGo Zero
AlphaZero (AZ) is a more generalized variant of the AlphaGo Zero (AGZ) algorithm, and is able to play shogi and chess as well as Go. Differences between AZ and AGZ include:[1]
AZ has hard-coded rules for setting search hyperparameters.
The neural network is now updated continually.
Go (unlike Chess) is symmetric under certain reflections and rotations; AlphaGo Zero was programmed to take advantage of these symmetries. AlphaZero is not.
Chess can end in a draw unlike Go; therefore AlphaZero can take into account the possibility of a drawn game.
AlphaZero vs. Stockfish and elmo
...
►Kingscrusher chess resources: www.chessworld....
►Kingscrusher's "Crushing the King" video course with GM Igor Smirnov: chess-teacher.c...
►FREE online turn-style chess at www.chessworld....
goo.gl/7HJcDq
►Kingscrusher resources: www.chessworld....
►Playlists: goo.gl/FxpqEH
►Play and follow broadcasts at Chess24: chess24.com/pr...
#KCChess
#KCChess
zero
♞ Challenge KC and others for turn style chess at bit.ly/chessworld
♚COURSES kingscrusher.t...

[@kingscrusher]

Replayable game with indented variations: www.chessworld.net/chessclubs/ltpgnviewer32/ltpgnboard.asp?GameID=5023622&v=NTTRpmUFWh8

[@RobBCactive]

kingscrusher Awesome game, Leela sac-ed and reacted with play that shifted the true centre to h2-h4-f7-f4!!! This should be called Leela's poisoned rook

[@LouisBarjon]

true pgn : 1.e4 e6 2.d4 d5 3.e5 c5 4.c3 Nc6 5.Nf3 Nge7 6.dxc5 Ng6 7.Be3 Ngxe5 8.Nxe5 Nxe5 9.f4 Nd7 10.b4 b6 11.c6 Nf6 12.Qa4 a6 13.Na3 Bd6 14.c7+ Qd7 15.Bb5 axb5 16.Qxa8 O-O 17.Bxb6 Ne4 18.Qa7 Bxf4 19.O-O g5 20.Rae1 f6 21.g3 Kh8 22.Re2 Rg8 23.Rc2 e5 24.Rg2 h5 25. Qb8 h4 26.gxf4 h3 27.Rc2 gxf4+ 28.Kh1 Qf5 29.Qxc8 Rxc8 30.Re2 Nxc3 31.Ref2 d4 0-1

[@RobBCactive]

Paul Thompson Actually I had a game rather like that and I'm a very error prone fallible player. It's fairly standard to push on one side but then switch the attack or balance prophylaxis on one wing, against attacks on opposite wings.

[@imroseuddin4680]

I wish i could give this video 2 likes. 1 for Leela and 1 for your hillarious but pure laugh.

[@paulcazier4819]

I've been a fan for many years now and I must say that you are clearly one of the very best, if not the best, chess youtuber out there. Thank you so much for such great content.

[@kingscrusher]

Well my enthusiasm is linked to the games in question and these are amazing games to go over - thanks :)

[@AndrewBackhouse1]

Paul Cazier Agreed

[@StringDogg]

Sci-Fi Thorn Pawn? Spectator pieces watching dolphines in a Siberian prison? G5??? This has to be the funniest and most enjoyable chess video I've seen :D Great work Tryfon this is top-content, fantastic game

[@Durian_Defense]

When chess players play for draws in boring games, we're scraping the barrel for creative and artistic chess content. When Leela plays Sci-fi chess like this, its like we find an enormous lake of chess creativity and artistry that is absolutely teeming with life. After hundreds of years of chess, just when you thought everything's been studied and played, suddenly It's like a ... breath of fresh air!

[@martinkaufmann4067]

I saw this video twice in a row and I cannot stop laughing. If I could give 100 likes I would do it. On one side this is chess of another planet and on the other side kingscrusher is amazing. Thank you, Tryfon.

[@palfers1]

As soon as the oppo king position has settled, Leela seems to lurch into action, never taking her eye off the oppo king. She loves locking up oppo pieces and she loves marching her pawns up the board. It's a strategy we see repeatedly from Leela. She's able to bring such aggression against the oppo king that it really doesn't seem to matter what's happening on the other side of the board. Perhaps that's over-simplistic, but it's a joy to watch, whatever it is!

[@Christoff070]

She times her attacks so well. And this is against some of the best chess engines! Forcing them to sac a queen with her ability to think of lines where an advanced pawn is worth more than a minor or major piece!

[@remussayed1007]

I can't stop laughing after the queen sac.

[@Wakssbm]

The day we will stop asking stockfish what's the best next move is coming. Slowly, but this day is coming with certitude

[@richardl6132]

slowly?

[@guepardiez]

I never thought I would laught out loud watching a chess game. Fascinating stuff.

[@ElectroplatedToaster]

Your enthusiasm in this review is heartwarming. I was laughing _with_ you! And, btw, great analysis as always, keep it up! And also, thanks to the people submitting these lovely games.

[@Traumtheater0]

Wow! I just realized you made a whole playlist of Leela-Gams. And there arealmost 100 games already! AND I watched them all so far:-) Great job, KC!

[@GnuReligion]

Wow, you are choking on emotion on this game. I think you are in love with Leela.

[@kingscrusher]

Unfortunately if you see the film "Her", AI is likely to "600-time" you if you fall in love with them. I will have to stick with my human loves for the time being and just appreciate her Chess :)

[@GnuReligion]

Well, I have watched "Her" (and like The Breeders "Off You"), but do not know what "600-time" means. Still, though you are obsessed with Leela, am still following you. Seems that the digital chess era is due for a merger between Leela and Stockfish, so that when one has the most definite move, that one will be selected.

[@robharwood3538]

I haven't seen Her yet (but will now!) but I think he's amplifying the expression, to 'two-time' someone, which means to (romantically) cheat on someone with another person. Presumably 600-time means to cheat on a person with 599 other people, I'm imagining.

[@GnuReligion]

Oh, OK, I get it now! Maybe I am just not the jealous type.

[@juggernaut316]

Absolutely hilarious game! I could not stop laughing at the queen emnprisoned in Siberia. It seems Leela has finally combined her pawn pushing prowess with the strategic positional sacs of Alpha Zero.

[@adnan4688]

I am so happy to see you that happy man,lol.Its amazing,and i feel excited like that also

[@u.v.s.5583]

Boy, you had so much fun, and so did we!

[@bhgtree]

Thanks KC, These games are so enjoyable, I'm can't wait each day to check and see if you've uploaded more.

[@Th3Wick3dOn3YT]

Those pieces were in a prison in Siberia. Ridiculous play, at one point just down a whole rook and able to keep activity. Fascinating game. Thank for Sharing.

[@marekzawal1744]

One of the best Kingscrusher's commentary ever. For me. So much passion and emotions, Leela broke the system, chess is complex on cosmic level.

[@aleratz]

Funniest video on Leela till now!

[@mateipopescu4338]

your enthusiasm made my day

[@Vennotius]

"Sci-fi thorn pawn" 🤣🤣🤣

[@AdrenalineVideos1337]

This game is a great example of the difference between neural networks and "traditional" chess engines

[@war6193]

Best analysis ever.

[@kobi2187]

I don't think it's any kind of higher understanding, but a type of experience and memory for which positions statistically lead to better boards. so in a way, we can potentially ask Leela why do you favour this position, and she'll say 87% from 83% based on this area of the board, for example. it's still a black box even if we get to directly ask why. the principles we infer can improve our play for sure though.

[@robharwood3538]

What a hilariously delightful reaction, KC! I couldn't help but join in with the laughing! 🤣

[@luciengrondin5802]

TL,DW: a good old thorn pawn is worth more than a queen in a gulag

[@pop9095]

Is he saying thorn? Sounds to me for all the world like fawn. Makes a lot more sense as thorn. Thanks.

[@beri4138]

pop9095 He does mean thorn, but he constantly mispronounces it.

[@robharwood3538]

It's just his regional accent, I think.

[@ludgerkres.1437]

Honestly, I was trying to hold in the laughter. I was successful. I almost burst at H3. But I stayed stoic. But I cannot help but laugh really hard when a top engine sacks a Queen in desperation. It is literally the most hilarious thing I see in these Leela Chess games. What makes a chess engine run through tens of thousands of positions, and to come up with "I sack the Queen now, MAYBE the pain will go away."

[@spinninglink]

It occurs to me that we've forgotten what the point of the game is. Checkmate the king. That's it. Everything else is irrelevant, unless it stops you from reaching that goal! Leela is here to remind us. Amazing analysis! Keep it up.

[@hochhoch-pu2gd]

Dear Kingcrusher, your analysis are amazing and fun to see. You are the only one covering so much of LC0 games that the question arise: what are the “new” general rules for playing these positional and tactical chess games? It seems, that My System was just the beginning of understanding chess better. With your M. Stean-like ability to explain things I’m curious, if you can transform the findings from these games in a reviewed or updated “version” of My System. I will be first subscriber of such a new-way chess book of yours. Thanks for all the efforts you spend for us in becomming better chess players!

[@maxpheby7287]

Alpha go did the same thing to stockfish in there games by constantly trapping the light squared bishop. However Leela seems to take this to a whole new level. Leela seems to "know" that it can trick the engine into a fatal piece placement by offering what the engine thinks is a good gambit but then turns out to be a trap. I think this a the difference between pattern recognition based chess vs calculation based chess.

[@user-hh2is9kg9j]

facilitating game, I am glad you are having fun making those videos as we are viewing them.

[@beri4138]

last shadow I don't think facilitating means what you think it means.

[@user-hh2is9kg9j]

LOL i meant fascinating. it was auto-corrected wrongly.

[@fujiapple9675]

8:42 Kingscrusher bracing for what he is about to reveal.

[@chardonnay5767]

That’s some passionate commentary, which makes the game even more interesting.

[@Christodoulosk]

This was an awesome game!!! Thanks KC!

[@franklippert4278]

No Leela game is complete without 'fawn' pawn.

[@marcusklaas4088]

Beautifully entertaining game, superb commentary. Thank you. Subbed.

[@manuelgarrido5602]

Leela is playing as a real genious stratege able of create disjonction between material and "spiritual"... Trapping materialistic focus to superior understanding.... Marvellous to watch....

[@adazacom]

This reminds me of a game in which Alphazero kept half of Stockfish's pieces bottled up for most of the game. To me it feels like the NN engines are playing like a cat with a mouse. Do you think that traditional engines have been taught how roughly to value pieces like kids are: Queen is 9, Rook is 5... Pawn on the 7th is... and NN engines have had to come up with their own piece rating system and what they've come up with is one entirely dependent on mobility? Poor Komodo is tightly holding onto this pawn like a treasure chest dragging it to the bottom of the sea. It is interesting to me that apparently Leela (and A0 before her) are choosing moves that encourage the traditional engines to do other things than spring themselves from their self-inflicted trap.

[@modolief]

I wonder what Kasparov would think of this game. He was really good at leaving pieces hanging all over the board, or making little dink moves during the middle of an attack. This is exactly the style of play in the present game.

[@4merxtian432]

Leela does all this basically down a rook the whole game

[@richardthomas1071]

Love your Leeela games ,great youtube channel

[@robertoh.20]

Leela playing god-level chess games.

[@jeremykothe2847]

Getting the feeling that Magnus is going to be the last human player that people wow over. Up until now we have taught and corrected the engines - in super-slow-motion so we can keep up, but it was our thinking and logic applied rapidly. And because it's not relying on 27-deep tactical traps but instead playing on principles, *we can follow it*.

[@gJonii]

Jeremy Kothe at least in go, alphazero has intuition that far exceeds that of humans. Humans try to figure out what are the principles it follows, but no one quite understands and it's just a deeper level of the game we're observing.

[@manuelgarrido5602]

Yes.... It seems that they have access to more powerfull intelligence that us humans... So they can understand principals we don t .... ........... And is this speaking much about our futur ?

[@robharwood3538]

Leela Zero (the Go version of this AI; precursor to Leela Chess Zero) is making steady progress in the same/similar directions as AlphaGo Zero. It's only a matter of time and computing resources until it gets to the same level. (Folks, you can contribute to both these projects, LZ, and LCZ, by running their distributed training play programs. Google 'em up if you want to help.) 😊

[@devvanbutler2758]

This is....wow

[@alephnull4044]

It is a pretty comical situation what you described with White's spectators in the other corner of the board haha.

[@dr.mikeybee]

So instructive! Thanks.

[@chessdad182]

That was an insightful position. It seems like the material value of the pieces outweighs the value of mobility to Komodo.

[@TheMarksT]

That reminds me of a game I just played on chessdotcom. There I needed to blunder to loose, in a couple of ways too, as my Queen was not in Siberia, but a Rook and 2 (recently) connected passed pawns were, with no blockaid and liberation at hand for the one about to Queen. I underestimated the position, and was suddenly mated-- being nagged by a forearm slightly out of place (snapped it back into place later -- had a fall on it once), explaining why I missed the hanging Knight (I actually made the right move and took it back before confirming). It also would have been a better fight had I just moved my Queen to g3 to stop the mate, maybe even a win had I taken the right approach to prevent the mate like that (but a fork taking the pawn on the 7th rank existed too). So even better on my 29th move just moved my Knight to g3 to counter check (28...Qg4+) that I lost the Knight on too and put my King in danger, after an exchange that half opened the g file and subsequently after moving my B-g2 instead to block the check, not noticing my Knight was hanging on e2. Otherwise it would have been a great win agaisnt a very strong player.

[@InfiniteUniverse88]

Putting "tactics in context" is strategy, not tactics. Leela's strategy was to divide and conquer.

[@devvanbutler2758]

What did I just watch? Im actually gagging on this game

[@extollo]

fun commentary. the c7 pawn must have skewed the evaluation leaving those defending pieces locked down. perhaps komodo would have abandoned it at a longer TC

[@fujiapple9675]

Leela incorporated the French Defence, Advance Siberian Variation.

[@luisfelipehserrano6176]

At least those two pieces had a nice view. The poor knight on a3 moved once and never again.

[@paultnt1]

Thanks KC these games show the neural network is stronger than p powerful engines - what rating is leela and can you give her a game plz

[@MrNekket]

Greetings from Siberia 😂 Amazing video. Fantastic game. Thank you, Tryfon.

[@Joelk76]

Love leela games. Kingcrusher what is the total score between leela and stockfish 7,8,9 and comodo?

[@4merxtian432]

I run some games on Twitch under Edosani. SF and Komodo still have a slight edge I think but there are games Leela plays so well from beginning to end and the brute force engines don't see they are losing until way too late. In fact it even looks to humans like Leela isn't winning either but she is.

[@gregainsborough9866]

Great Video!

[@modolief]

Great commentary !!!

[@rotcod2886]

Good stuff. Don't worry so much about explaining the Thorn pronunciation Tryfon. Let other people in the comments answer. If people don't understand accents, that's their problem. I'm fascinated by them. Why don't we say Paree instead of Paris? We easily could! I'm not sure what linguists would say. But I think it would be asking us to speak in a foreign accent, and that's just unnatural enough to be, what, uncomfortable? Inconvenient? Flat out wrong? Anyway, if you want, I guess you could put it in the info, but screw it. Just let people figure it out, I say.

[@spodule6000]

I always laugh and think Phorn Porn !

[@PaulFurber]

Gulag would seem to be the right term here. Great analysis.

[@modolief]

15:09 -- You ask: "The big question: do we really think the traditional engines are strong tactically when they can play these tactical sequences which win material but leave their pieces terribly irrelevant?" I can offer some ideas towards an answer. I've looked over _Mastering Chess and Shogi by Self-Play with a General Reinforcement Learning Algorithm_ by Silver et. al., the relevant paper by the AlphaZero team. That version of AlphaZero chess did lose some games to Stockfish 8, so it's play wasn't perfect (though it drew or won most of the games). Clearly the AlphaZero neural network approach is much more fruitful for Go than it is for Chess, but, as we have seen, it is quite effective for chess as well. There might be some alpha-beta optimizations that could go into a neural network centric program. In the section on methods the authors comment "It is likely that some of these techniques could further improve the performance of AlphaZero"; but their main goal was to show that a neural network plus MCTS approach to chess could be highly effective so they didn't go for a fully optimized chess engine. I'll quote the first few paragraphs of the methods section because it also gives a nice definition of a "traditional" chess engine, as compared to simply saying "brute force"; and it also shows the above comment about "further improve the performance of AlphaZero" in context, so that we might have an inking of how stronger tactical programming might enhance Leela Zero. Here's from the paper: Methods -- Anatomy of a Computer Chess Program -- In this section we describe the components of a typical computer chess program, focusing specifically on Stockfish (25), an open source program that won the 2016 TCEC computer chess championship. For an overview of standard methods, see (23). Each position s is described by a sparse vector of handcrafted features φ(s), including midgame/endgame-specific material point values, material imbalance tables, piece-square tables, mobility and trapped pieces, pawn structure, king safety, outposts, bishop pair, and other miscellaneous evaluation patterns. Each feature φi is assigned, by a combination of manual and automatic tuning, a corresponding weight wi and the position is evaluated by a linear combination v(s, w) = φ(s) >w. However, this raw evaluation is only considered accurate for positions that are “quiet”, with no unresolved captures or checks. A domain-specialised quiescence search is used to resolve ongoing tactical situations before the evaluation function is applied. The final evaluation of a position s is computed by a minimax search that evaluates each leaf using a quiescence search. Alpha-beta pruning is used to safely cut any branch that is provably dominated by another variation. Additional cuts are achieved using aspiration windows and principal variation search. Other pruning strategies include null move pruning (which assumes a pass move should be worse than any variation, in positions that are unlikely to be in zugzwang, as determined by simple heuristics), futility pruning (which assumes knowledge of the maximum possible change in evaluation), and other domain-dependent pruning rules (which assume knowledge of the value of captured pieces). The search is focused on promising variations both by extending the search depth of promising variations, and by reducing the search depth of unpromising variations based on heuristics like history, static-exchange evaluation (SEE), and moving piece type. Extensions are based on domain-independent rules that identify singular moves with no sensible alternative, and domaindependent rules, such as extending check moves. Reductions, such as late move reductions, are based heavily on domain knowledge. The efficiency of alpha-beta search depends critically upon the order in which moves are considered. Moves are therefore ordered by iterative deepening (using a shallower search to order moves for a deeper search). In addition, a combination of domain-independent move ordering heuristics, such as killer heuristic, history heuristic, counter-move heuristic, and also domain-dependent knowledge based on captures (SEE) and potential captures (MVV/LVA). A transposition table facilitates the reuse of values and move orders when the same position is reached by multiple paths. A carefully tuned opening book is used to select moves at the start of the game. An endgame tablebase, precalculated by exhaustive retrograde analysis of endgame positions, provides the optimal move in all positions with six and sometimes seven pieces or less. Other strong chess programs, and also earlier programs such as Deep Blue, have used very similar architectures (9,23) including the majority of the components described above, although important details vary considerably. None of the techniques described in this section are used by AlphaZero. It is likely that some of these techniques could further improve the performance of AlphaZero; however, we have focused on a pure self-play reinforcement learning approach and leave these extensions for future research.

[@jeremykothe2847]

It seems obvious that "fine-tuning by hand" the search algorithm using similar heuristics to traditional engines would give a boost. It should also be obvious that it is completely unnecessary except as an optimisation. The difference between a NN eval and a traditional one is that the NN eval contains the results of actual prior searches. Merely continuing the iterations will end up learning those optimisations inside the network itself. Now of course a general-purpose net will not be as efficient as a handcrafted algorithm, but the cost of handcrafting one that is better will approach infinity as the cost of the NN approach goes to 0.

[@konzy2]

Who told him how to say the graphics card name properly, I liked how he said One Oh Sixty

[@juggernaut316]

I thought it was endearing.

[@kbear8293]

Siberian prisoners on holiday watching dolphins

[@subudjj9368]

Whats the general ratio of leela wins to losses against traditional engines?

[@kingscrusher]

it really is irrelevant to me because Leela did something unthinkable recently - on two normal Graphics cards, she essentially draw with the computer world champion and other major engines who had massive hardware. The intuition test games massively favour traditional brute force engines and Stockfish 8 definitely still has the edge at the time of this video. But this can change just within potentially weeks or months with current evolution progress - especially promising is the new Test server.

[@klirsnichy]

Amazing video :D

[@abdelazizZ]

I like your analysing way specialy bots games Keep going

[@TheAfroNoah]

FANTASTIC absolutely fantastic

[@Grzegorz54321]

I watched this game and only said WTF?! :D

[@h1a8]

Thank you for the excellent video! Was this game played on your machine? If so then what’s your configuration (processor, game card, game time , etc)?

[@kingscrusher]

David Grovesner's machine - see pinned comment for interactive PGN which mentions more details

[@russellmemo1353]

I just passed by to say : Haew Mr Kingscrusher

[@kasuha]

Technically, it all boils down to the fact that Komodo couldn't see the attack to the end while Leela obviously did see it already while it was allowing the queen to take the rook. I would guess Leela and possibly other NN engines to come will bring a lot of inspiration for improvements of traditional engine position evaluation and the end result will be somewhere in between, engines that are much stronger in position evaluation but capable of running it faster and more reliably than neural networks.

[@manuelgarrido5602]

Kasuha Interesting point of view... Why do you think brut force ingines will still be superior to NN engines when they will be implemented by better evaluation ? May be NN will have no real opponent ...

[@kasuha]

Both Komodo and Leela are "brute force engines". Leela just wins because it can select better moves to investigate and that leverages the fact that it only manages to investigate less of them. It still likely goes to hundreds or even thousands per move in this match (while Komodo likely investigated tens to hundreds thousands moves in the same time) I am not 100% sure Leela saw the end of the line but I see it as very likely since the intuition is only part of Leela's play and without calculating hundreds of positions ahead it would lose.

[@jeremykothe2847]

Here's the reason that won't happen, Kashua. Given infinite human resources, a combination of approaches, yes, would yield the best results. But given *zero* human resources and some time, a net will get there anyway, then keep going. It's only in the short term that we can contribute. As with image recognition before it, people have just given up completely on heuristics because the nets can provide their own which are actually performant by definition. We cannot compete. The good news if you're paranoid is that this is the bleeding edge. But give it a few years and... let's just say we're all going to have a lot of spare time.

[@jeremykothe2847]

Apologies, @kasuha, I am terrible with names.

[@kasuha]

A chess engine only needs to be strong enough to never lose. I think we're not too far off that point, the Stockfish vs AlphaZero matches were rather tight and current Stockfish is even stronger. Just a little added value might get it there and then any further effort with NN is pointless in terms of superiority, they only can give us some new play styles. Note also that NN training does not come for free, you need actual people to spend actual effort to set it up, run it, and pay for the power spent. I would say the costs associated with training a good NN are quite on par with costs associated with designing a good traditional chess engine.

[@Euquila]

Wow... had to watch this one twice

[@paultnt1]

KC in your analysis you snd SF only found a draw continaution in one of the lines but leela would probably have found a win in her neural network

[@AndrewBackhouse1]

I sometimes play this line as black. I hope one day to produce a masterpiece like Leela.

[@richardsleep2045]

Nice one, as everybody is saying

[@evgenys177]

You are A M A Z I N G !!!!

[@romankusnir45958]

Well, all we need is imagination.

[@drob9673]

it's another crazy Leela game so first I like the video then I watch it :D

[@zsoltbihary3347]

Very interesting and funny.

[@kobi2187]

there are chess lessons here, in the most fundamental aspects of what to emphasize in your tactics. by making them irrelevant, it's as if they were captured. The engine probably gave too much score to the c7 pawn which was almost a new queen, to move the bishop and queen out of there. But in essence, there wasn't a real trap, komodo could have taken the bishop and then the queen out. is my idea wrong here?

[@Christoff070]

Nah I like your points, except the last part about getting the queen and bishop out. With the bishop blocking the queens direct diagonal to the king side, it would have taken at least two moves if not three, thanks to the black Knight at e4, to get the Queen back to defends, which is way too long against Leela sci-fi cavewoman!!

[@modolief]

I hope this doesn't sound trite, but this game reminded me of "Neo vs. Agents" -- this scene: kzbin.info/www/bejne/mom4hpmwr5V3otE -- even though they are "upgrades" and put up some resistance, the image I have in my mind is this sort of aikido way of holding onto the arm of one opponent and using him as a convenient shield or baffle against another opponent, while targeting the third opponent, and then selectively taking each one out with a killer blow.

[@LouisBarjon]

true pgn : 1.e4 e6 2.d4 d5 3.e5 c5 4.c3 Nc6 5.Nf3 Nge7 6.dxc5 Ng6 7.Be3 Ngxe5 8.Nxe5 Nxe5 9.f4 Nd7 10.b4 b6 11.c6 Nf6 12.Qa4 a6 13.Na3 Bd6 14.c7+ Qd7 15.Bb5 axb5 16.Qxa8 O-O 17.Bxb6 Ne4 18.Qa7 Bxf4 19.O-O g5 20.Rae1 f6 21.g3 Kh8 22.Re2 Rg8 23.Rc2 e5 24.Rg2 h5 25. Qb8 h4 26.gxf4 h3 27.Rc2 gxf4+ 28.Kh1 Qf5 29.Qxc8 Rxc8 30.Re2 Nxc3 31.Ref2 d4 0-1

[@futurefox128]

Nice game, but not so great when you analyse it yourself and immediately see that it's full of little mistakes from both sides. :( White's position was fine and probably even slightly better at some points, but both played some (very) inaccurate moves and 25. Qg8 was a huge blunder (getting the Queen even deeper into the prison), that immediately loses the game. From a strategic point of view, white could have pulled it's pieces out by playing moves like Be3 at some point. It seems Komodo 9.02 really prioritizes hanging on to every bit of material instead reactivating its pieces, also being completely blind to Leela's king side attack. This becomes evident when K played moves like Rc2 just to needlessly defend c3, costing white precious time. K should have played Rg2 immediately, which it did just 1 move later, showingcasing that Rc2 was indeed pointless and a waste of time. Thanks for the upload though! It was an enterntaining game nonetheless. :)

[@greatwolf.]

Yes, ten-sixty thank you for saying it right ;)

[@jeffreygolding341]

Ho has Leela done against Alpha Zero?

[@dannygjk]

AZ retired after destroying the great pretender.

[@paultnt1]

AZ would have destroyed leela it was playing on the most powerful system available however as leela learns like a human within time leela will be stronger

[@Nuhyamin1]

Komodo just fell for Leela.

[@sausage4mash]

that game would of got Tal scratching his head, I mean at 3:09 who'd have black here ?! who would say oh yeah black is clearly better ,lols .

[@beri4138]

Lord of the Pies But is black better here? What makes you so sure? I'm not even sure how to answer that. Can I trust a Stockfish 9 analysis of this position? Or will it fail to grasp it in the same way Komodo 9.02 did?

[@beri4138]

Lord of the Pies But is black better here? What makes you so sure? I'm not even sure how to answer that. Can I trust a Stockfish 9 analysis of this position? Or will it fail to grasp it in the same way Komodo 9.02 did?

[@abdelkrimzaouali4256]

Leela simply make me disgust of human chess. Leela simply killed all human immortal games.

[@kingscrusher]

I think the opposite is true - Leela will encourage more amazing games - I am already experiencing the effects of aiming for positions with connected passed pawns etc in my own play - and it is exhilarating to have Leela as my new virtual coach.

[@dannygjk]

Typical... brute force is led by it's nose by greed.