It wouldn't work well unless evaluation is given by those cloud engines, that actually play the game. Browser or local engines are obviously too weak to provide accurate analysis
@@null1700I think that's the point. That sort of information should be logged and published as the games are being played. I reckon you could compare them and see some interesting results
When I first started watching him, and was like a chess beginner, I found it kinda pointless, annoying, waste of time, but now I find it interesting how it could've played out and it actually explains why certain moves are better than others (where you have to think many moves ahead) so yeah it's more interesting like that
It's absolutely staggering to see Leela play this game. Stockfish plays like an absolute bloodthirsty demon but LC0 just Sun Zu's through it relentlessly. I am far too smooth brained to understand the plays but it's gorgeous to watch
To be fair every game in this video is LC0 playing white from a position that is slightly winning from the opening. Stockfish would win these positions just as easily if not easier (most likely) Still cool to watch, but it's not some nail in the coffin of Stockfish
@@jordanmungo917 were they "slightly winning from the opening" tho? the second one had leela losing a bishop at the start. white generally has a very tiny advantage; but i don't think that microscopic advantage matters that much among the top AIs, especially when they are programmed to do theory moves in the opening. but yea, definitely far from stockfish being donezo... yet. what would be amazing tho... is if stockfish actually learns from these losses and starts dreaming up counters for that h6 pawn from leela. it would be spooky if we can actually track stockfish's evolution in terms of chess theory... what if stockfish actually stops playing like an aggro demon at the next tournament and starts positioning defensively/long-term like leela did here?
@@walidali8669 So what tournament did Leela win then? Superfinal: Stockfish won 52-48. Cup 12: Stockfish won 14.5-13.5. Swiss: Stockfish won at 15.5 points, Leela second at 14.5 points. FRD league is currently running, might be where Leela actually can win for once. Stockfish shouldn´t be this dominant if what you say is true.
Something I'm learning from these engines. Its not just about gaining space. It's denying space. In that second game Lela completely blocked Black from its side of the board and patiently pushed its way closer to the other side while preventing black from getting past its front lines.
@@michaelblankenau6598 usually gaining space would mean squares on the opponents side of the board and protecting your side you would just call defense or prophylaxis or something like that
@@michaelblankenau6598 I mean yes, obviously...if you look at it as total squares captured and controlled. But I was more looking at it from the standpoint that moving forward and gaining squares can be done while allowing infiltration into your own side of the board. But if you deny access to your side AND move forth and gain more space, this is what the engines do.
Would love to see how Stockfish handled the same positions as white. Would be really interesting to see where the two computers deviate from one another in these paired games and how that ends up affecting the result of each match.
Your AlphaZero vs Stockfish breakdowns got me into your channel and into chess in general. You are very good at commentating chess games and making any game interesting
Really immersive video Gotham. Played this on my TV and I felt like you were right there in the room with me like a little human in the TV explaining chess to me
Well, for me it has fallen after going neural network...just after Leela won TCEC. It has to be noted that Stockfish is still the champ, but after going NN, it was just crushing opponents, including Leela. Seems the situation is getting re-balanced. I was expecting this thing to happen because Leela's approach in algorithm is better in the long run but its success is linked to the progress of computers.
Love when you do chess engine reviews, Levy! I'm an avid TCEC watcher, especially the long time control parts of seasons and bonus rounds. Love your content! Sending love from New Zealand!
Excellent video Levy. Love these engine deep dives. The different variations that you explain give a much better understanding of these complicated engine moves. Would love to see more videos like these in the future.
These engines are like chess avengers. Stockfish is the team leader, Leela is the calm and swift female character, Torch is the new kid in the team who comes from a rich family, always naughty, and super aggressive in fights.
This was such a fun video to watch! I don't usually care so much about chess engines, but you've done such a good job explaining why they're important to the evolution of the game PLUS weaving a narrative around this competition... dramatic and engrossing!
I would love to see content like this with both eval bars. It is interesting to see the stockfish eval bar be like "it's a draw" and the a few moves later be like "just kidding I'm losing". Curious to see what the other eval was at the same time.
That second game was very Magnesian, all pieces out, all pawns reluctant to move until the exact right time. Amazing. This is like watching Hikaru adopting 2750s.
It’s funny how much having an eval bar has become standard to look at as a source of truth, but with bot chess you realize you don’t actually know what the evaluation is cause the bots disagree
So, basically, Stockfish's aggression decimates any human players, but Leela's perfect defense negates its advantage. And thus, the unstoppable force can't touch the immovable object.
Current GPT still doesn't really *understand* chess. It can just get further into a game before losing track of everything and dying. Unless there's some kind of new breakthrough in the way GPT fundamentally operates I doubt it will match GMs, let alone Stockfish.
GPT 3.5 instruct plays well Allegedly, it has a strength of 1800 elo I wonder how GPT 4 instruct will perform considering how GPT spT4 chat gave Levy a run for his money
I love these chess engine videos and I would love for you to do more. It would also be fascinating to hear you kind of talk about the different theories that go into building these chess engines because Stockfish and Leela are built on totally different principles. I find that fascinating as well, but please more of this.
LCZero foiling the castle essentially took two pieces for one, the pawn taken and the h file rook stuffed in the corner for several more moves, unable to move to the center files
This, its still considered #1 engine but half the comments are convinced A0, Lc0 or Torch are better solely because Levy only shows the games where said shakeups occur and not the ones were SF16 expectedly wins.
Don't get the wrong idea, Stockfish is higher rated and wins a greater amount of games against Leela. Also the openings featured in these games are notoriously bad for black at computer level. Stockfish vs Stockfish, Leela vs Leela, Stockfish vs Leela and Leela vs Stockfish all would result in a high probability of white winning the games. Don't compare Stockfish to an IM. :(
As each position is repeated on both sides, It could be interesting to analyze difference in gameplans. Did both engines develop the same ideas towards victory? Especialy when there is a difference in the outcome (as it was the case for game 3). Why did Leela won as white but stockfish didnt win when it had the white pieces in the same position. Leela lost the superfi but it's getting closer
Interesting, that originally, AlphaZero was extremely aggressive, yet LC as a "spiritual successor" is much more ... "calm and patient" in a way. - And Stockfish flipped in the opposite direction, perhaps?
That makes sense, since Stockfish is a classical engine which has to "prove" that a move is good during its search in order for a move to be chosen. Meanwhile Lc0 can just rely on intuition and play a move knowing it will probably lead to an advantage at some point, even without having calculated it to the end. And I think the reason AlphaZero looked aggressive is just because Stockfish 8 played flawed moves, which it mercilessly punished. If you put Lc0 against the old Stockfish 8, I'm sure it would look similar.
I like watching the computer videos because it shows me deep middle game and end game ideologies. Its like seeing the frontier of chess and where it can be...