Could you maybe compare how the engines play each game in a pair? I feel it could be really interesting to compare see how their approaches differ in the same opening, and how that results in different outcomes
What I would really like to see is the opposite game, how Stockfisch drew the same opening against lila, like a quick highlighting of the different choices the engines did
For the record, Leela essentially is AlphaZero. It uses the same network architecture (with some minor optimizations), the same tree search algorithm, and the same training methodology. So unless AlphaZero saw some significant changes the two bots would be identical strength. Likely Leela would be a touch stronger, as it has been trained more.
@@afuzzycreature8387 Stockfish is the best engine by far. It uses a super clever input structure that was adopted from Shogi networks. The network takes a fraction of the input required for Leela and uses a more expansive (if less efficient) tree search algorithm. This means it calculates significantly more positions than Leela at any given depth when given the same amount of time.
False, while Leela is a modern more powerful engine. They’ve got nothing in common, not the same at all. In a way Leela is closer to SF than it is to Alpha zero.
I dont care who is going to win, I love your recaps of these computer vs. computer games, but (as they play pairs) I would like to see on which move they disagree and what deviation was played in the other game.. thanks for your insights and for making clear what exactly is the starting position they were given :) For the second edition of your book, I suggest you state clearly that one cannot castle when in check - the explanation on page 8 (english version) is not perfect in this regard (either you or I missed that there) - rest of the book is great, would buy again :)
Chess is fucking crazy. I mean if you start you learn all the important things like develop your pieces, push pawns, not too many queen moves in the early game, no knights at the edges of the board etc. and this can get you to 2000+ elo if you are good and train very well. And then comes Stockfish and it is like: You know the rules, so I do, so I ignore the rules most of the time. Fuck those rules. Stockfish moves knight to the edge, does a lot of queen moves in early game, .... and finally wins the game
I'm fascinated by AI. I worked in the AI industry back in the '90's, long before the "big iron" of computing became so ubiquitous (and cheap) such that anyone with a Google login can do deep learning. Yes, I would enjoy more videos with Lc0. Thank you for the offer.
Isn't Alpha Zero way weaker than modern SF, since it beat SF8. Also isn't Lc0 just a superior implementation of the same methodology as Alpha Zero. Either way I suspect SF16 would crush Alpha Zero.
Would be interesting to see the evaluation from each side (not just the browser stockfish) But I would totally understand if that would be too much work ;) And I would also be interested in Leela's wins and I think it would be good to mention if the opening was won by both or only one engine.
I started to see a pattern in some of the moves we, as humans, dont find natural: If the opponent takes a piece, the bot evaluates whether there is a threat or if that piece can get away. If both of these dont apply the bot finds it more helping to develop first instead of retaking immediately
people (especially those of us below 1000 elo) will ALWAYS take back. It's just human nature. Even if you can calculate out 3-4-5 moves, taking back almost always makes sense, unless you can see it creates an immediate (next move) threat, taking back is just something people do. Being down material makes us nervous.
Hey! I'm watching this video in 2023! You're going to London? That's great! When are you coming to Pennsylvania? It'll be fun! You can guess my ELO. Apoiler: it's in the single digits, but I hope that improves after reading and practicing your excellent book.
Probably everybody knows already, but I am wondering, why don't RU-vidrs use the real Stockfish engine (45MB download) for evaluations instead of the Lite variant ... it's much more accurate and shows much more realistic eval. All stockfish and most Lc0 moves were that engine's top moves. Is it a deception so people would not see the real situation evaluation ? And then YT'rs say, that "even stockfish didnt see it" ?!?!?!
19:25 i like your pov regarding how stockfish is playing. Stockfish slowly maneuvers its pieces, finds a good spot, protects it, and then puts its attacking piece over there. Also, stockfish loves knight more than bishop or what????
Levy is a genius. At this rate eventually people will study enough of those games to play at this level. Levy perfoming analysis on all of these stockfish games is proof of him future proofing his channel. Hes in this for the long run and first person in history to make a full course for students in the future who are playing at 3600. Hes a genius!
I preordered a signed copy of the book but moved before it was released. I changed my mailing address but it was still shipped to the old house and when I reached out they basically said oh well.
*Sometimes i think about how if the universe is infinite, somewhere there is a race of super intelligent hive mind machines making Dyson Spheres around stars to gather enoigh energy to run a super chess computer because their biological creators only got that far in their use case before extinction*