TE
科技回声
首页24小时热榜最新最佳问答展示工作
GitHubTwitter
首页

科技回声

基于 Next.js 构建的科技新闻平台,提供全球科技新闻和讨论内容。

GitHubTwitter

首页

首页最新最佳问答展示工作

资源链接

HackerNews API原版 HackerNewsNext.js

© 2025 科技回声. 版权所有。

Positions chess engines don't understand

407 点作者 diplodocusaur大约 4 年前

17 条评论

sillysaurusx大约 4 年前
When OpenAI trained their Dota model to beat people 1v1 mid SF, they had a contest at The International (the big yearly dota tournament, RIP) like &quot;can you beat OpenAI&#x27;s new bot?&quot; and had a big bucket of prizes for anyone that could.<p>By the end of the night, the bucket was empty. People learned to cheese the bot by running &quot;out of bounds&quot;, so to speak -- normally in a 1v1, you&#x27;re supposed to stay close to your opponent, since they&#x27;ll be getting stronger if you leave. But the bot didn&#x27;t know how to deal with it when you snuck behind it (normally an insane maneuver, almost guaranteed to cost you the game in normal play) and prevented his army (called &quot;creeps&quot;) from running to the middle lane. (All you have to do is go wave your hand at them, and they mindlessly chase you.)<p>People did that over and over, and your own army would eventually overwhelm the bot and win. :)
评论 #27190342 未加载
评论 #27191695 未加载
评论 #27190422 未加载
评论 #27193641 未加载
评论 #27190812 未加载
评论 #27190539 未加载
airza大约 4 年前
An astonishing insert:<p>&gt;A good example of human exploitation of the engine&#x27;s failings is GM Hikaru Nakamura&#x27;s defeat of Rybka in the following three-minute blitz game from the Internet Chess Club. Nakamura cleverly locks the position so that progress is impossible for Rybka, then he offers two exchange sacrifices to the engine. With the position locked, the engine&#x27;s rooks have no value, but the engine thinks it has a winning material advantage. With a draw by the 50-move rule approaching, the engine sacrifices a pawn to avoid a draw, but this proves a huge mistake as Nakamura is then able to win the game, an incredible achievement!
评论 #27189770 未加载
评论 #27193669 未加载
评论 #27190515 未加载
评论 #27189852 未加载
matsemann大约 4 年前
When watching the WC games, I&#x27;ve seen it happen that a move wasn&#x27;t considered as a top move by the engine, but once played the engine realizes it&#x27;s actually crushing. Something about the heuristics used to prune the vast search space can make it miss sacrifices or seemingly sub-optimal moves that temporarily weakens the perceived position but has a huge payoff in the end. But humans find them. Of course, given enough time and depth the engine will eventually circle back and try the move. But it has no intuition.<p>Also, an engine without an endgame tablebase can be pretty stupid. There are certain rules one can deduct when there are few pieces left, but a min&#x2F;max engine will search forever, not knowing the patterns.
评论 #27189271 未加载
评论 #27189708 未加载
评论 #27189692 未加载
评论 #27189775 未加载
评论 #27189534 未加载
评论 #27191201 未加载
评论 #27193077 未加载
评论 #27193699 未加载
评论 #27192009 未加载
评论 #27189404 未加载
perihelions大约 4 年前
Here&#x27;s a particularly extreme example:<p><a href="https:&#x2F;&#x2F;old.reddit.com&#x2F;r&#x2F;chess&#x2F;comments&#x2F;ndz2lj&#x2F;simple_mate_in_93&#x2F;" rel="nofollow">https:&#x2F;&#x2F;old.reddit.com&#x2F;r&#x2F;chess&#x2F;comments&#x2F;ndz2lj&#x2F;simple_mate_i...</a><p>It&#x27;s a mate-in-93 puzzle that&#x27;s fairly accessible to humans, using abstract reasoning. But not chess engines. Comparing against the OP article, the main &quot;technique&quot;&#x2F;&quot;trick&quot; is zugzwang (#7), but on a dramatic scale.<p>I think this is the kind of position you could use to stress-test a candidate puzzle solver, just because of the shear size of the solution.
dwohnitmok大约 4 年前
Does anybody know if advanced chess&#x2F;centaur chess (chess play where a human uses a computer for assistance) is still a thing&#x2F;whether a human+computer combo is a meaningful improvement these days (i.e. last couple of years) over just a computer.<p>I can&#x27;t find any recent advanced chess tournaments and though I see quotes of people saying that the combo is stronger than a computer alone, I haven&#x27;t found any recent examples of a top tier engine by itself losing to a human + engine (e.g. Stockfish + human vs Stockfish).
评论 #27190486 未加载
评论 #27189299 未加载
rudi-c大约 4 年前
There&#x27;s a similar situation with Go where some positions utterly confuse bots trained on playing mostly normal games. There was this interesting research blog post on training a bot specifically to become good at solving one of these weird problems (Igo Hatsuyoron 120, the &quot;hardest go problem ever&quot;)<p><a href="https:&#x2F;&#x2F;blog.janestreet.com&#x2F;deep-learning-the-hardest-go-problem-in-the-world&#x2F;" rel="nofollow">https:&#x2F;&#x2F;blog.janestreet.com&#x2F;deep-learning-the-hardest-go-pro...</a>
评论 #27191426 未加载
mrslave大约 4 年前
Agadmator covered Kramnik v Leko 2002 (in 2018) titled &quot;Invisible to Engines | One Of The Greatest Moves Ever Played&quot; [0] which is worth a watch.<p>[0] <a href="https:&#x2F;&#x2F;www.youtube.com&#x2F;watch?v=yGnpewUKP88" rel="nofollow">https:&#x2F;&#x2F;www.youtube.com&#x2F;watch?v=yGnpewUKP88</a>
senkora大约 4 年前
It seems dubious to show that engines are sometimes bad at evaluating positions by giving a position with three black bishops on black squares.
评论 #27189209 未加载
评论 #27189256 未加载
评论 #27189451 未加载
评论 #27189285 未加载
评论 #27189378 未加载
microtherion大约 4 年前
The Hasek study in #2 is bugging me: Why can&#x27;t Black simply play 1. ... Rh8 ? It looks to me like this would gain the one crucial tempo to win the game. None of the discussion of the study I&#x27;ve found seems to consider this move (And the online analysis engines are useless, as they don&#x27;t properly understand that the original line is a draw, as the article notes).<p>ETA: Corrected typo, I originally wrote 1. ... Re8, which does not accomplish anything.
评论 #27190692 未加载
评论 #27193321 未加载
zone411大约 4 年前
So I actually checked the problems listed.<p>In the &quot;IQ Test #52&quot; position (FEN: 8&#x2F;1p1q1k2&#x2F;1Pp5&#x2F;p1Pp4&#x2F;P2Pp1p1&#x2F;4PpPp&#x2F;1N3P1P&#x2F;3B2K1 w - - 0 1) listed in #1 both LC0 and Stockfish play the correct line on my computer in seconds.<p>Both of them also play the right move Ba4+ in the second position of #1 &quot;William Rudolph vs.&quot; (FEN: 8&#x2F;1p1q1k2&#x2F;1Pp5&#x2F;p1Pp4&#x2F;P2Pp1p1&#x2F;4PpPp&#x2F;1N3P1P&#x2F;3B2K1 w - - 0 1) but they take quite a bit longer. Stockfish variants get it quicker.<p>Stockfish solves &quot;Hasek vs.&quot; (FEN: r7&#x2F;7k&#x2F;5R2&#x2F;p3p3&#x2F;Pp1pPp2&#x2F;1PpP1Pp1&#x2F;K1P3P1&#x2F;8 w - - 0 1) listed in #2 quickly. Both Stockfish and LC0 solve &quot;Lazard=F vs.&quot; (FEN: q7&#x2F;8&#x2F;2p5&#x2F;B2p2pp&#x2F;5pp1&#x2F;2N3k1&#x2F;6P1&#x2F;7K w - - 0 1) quickly.<p>Stockfish gets Bh3 from #3 &quot;Veselin Topalov (?) vs. Alexey Shirov&quot; (FEN: 8&#x2F;8&#x2F;4kpp1&#x2F;3p1b2&#x2F;p6P&#x2F;2B5&#x2F;6P1&#x2F;6K1 b - - 2 47) in seconds (7-piece end game tablebases installed).<p>The next position from #3 &quot;Spassky, Boris V vs. Byrne, R.&quot; (FEN: 3B4&#x2F;1r2p3&#x2F;r2p1p2&#x2F;bkp1P1p1&#x2F;1p1P1PPp&#x2F;p1P4P&#x2F;PPB1K3&#x2F;8 w - - 0 1) is easy for both Stockfish and LC0. They both get 50. c5!! right away.<p>Stockfish also gets the last position from #3 &quot;Stefan Brzozka vs. David Bronstein&quot; in seconds (FEN: 1r6&#x2F;4k3&#x2F;r2p2p1&#x2F;2pR1p1p&#x2F;2P1pP1P&#x2F;pPK1P1P1&#x2F;P7&#x2F;1B6 b - - 0 48) Rxb3+.<p>Stockfish and LC0 see Kd1 in &quot;Lamford=P vs.&quot; from #4 (FEN: 8&#x2F;8&#x2F;8&#x2F;1k3p2&#x2F;p1p1pPp1&#x2F;PpPpP1Pp&#x2F;1P1P3P&#x2F;QNK2NRR w - - 0 1) but believe Rg2 is also winning.<p>Stockfish gets c8N from &quot;Randviir=J vs.&quot; (FEN: 5nr1&#x2F;2Pp2pk&#x2F;3Pp1p1&#x2F;4P1P1&#x2F;6P1&#x2F;5K2&#x2F;8&#x2F;7n w - - 0 1) in #4 in about 2 minutes on my computer.<p>Stockfish gets Bf5 from &quot;Simkhovich=F vs.&quot; (FEN: 8&#x2F;8&#x2F;2pk4&#x2F;8&#x2F;p1p3B1&#x2F;PpP5&#x2F;1P6&#x2F;r1NK4 w - - 2 2) in #5 in seconds. LC0 also gets it.<p>The next two positions are mentioned as easier for programs and they are:<p>Qe3 from &quot;Deep Blue vs. Garry Kasparov&quot; in #6 (FEN: 1r6&#x2F;5kp1&#x2F;RqQb1p1p&#x2F;1p1PpP2&#x2F;1Pp1B3&#x2F;2P4P&#x2F;6P1&#x2F;5K2 b - - 14 45) is very easy for both Stockfish and LC0.<p>Both also get &quot;Vladimir Kramnik vs. Peter Leko&quot; (FEN: 6k1&#x2F;5p1p&#x2F;P1pb1nq1&#x2F;6p1&#x2F;3P4&#x2F;1BP2PP1&#x2F;1P1Nb2P&#x2F;R1B3K1 b - - 0 25) in #6 quickly.<p>&quot;Matous=M vs.&quot; (FEN: n2Bqk2&#x2F;5p1p&#x2F;Q4KP1&#x2F;p7&#x2F;8&#x2F;8&#x2F;8&#x2F;8 w - - 0 1) is indeed harder for Stockfish and LC0 than expected. I&#x27;ve confirmed mate in 13 in my mate solver.<p>In &quot;Nigel Short vs. Vladimir Kramnik&quot; (FEN: r3r1k1&#x2F;1bp1Bppp&#x2F;pb1p4&#x2F;1p6&#x2F;1P6&#x2F;1BP2P2&#x2F;P1P2PKP&#x2F;R3R3 b - - 6 19) from #9, the engines like ...a5 more than ...c6. Hard to say that this move doesn&#x27;t also win without more analysis.<p>Stockfish wants to play c6! and b4! from &quot;Marwitz=J vs.&quot; from #10 (FEN: 2K3k1&#x2F;1p6&#x2F;R3p1p1&#x2F;1rB1P1P1&#x2F;8&#x2F;8&#x2F;1Pb5&#x2F;8 w - - 0 1) right from the start. LC0 takes longer but gets it as well.<p>In &quot;Anish Giri vs. Maxim Rodshtein&quot; (FEN: 8&#x2F;5pkp&#x2F;3p1np1&#x2F;Rpr5&#x2F;8&#x2F;6P1&#x2F;PB3PKP&#x2F;8 w - - 6 34) from #10 both Stockfish and LC0 like 34. h4 over 34. a4. More analysis would be needed to see if h4 is not also winning.<p>Last position &quot;IQ Test #16&quot; (FEN: 5k2&#x2F;4bp2&#x2F;2B3p1&#x2F;1P4p1&#x2F;3R4&#x2F;3P2PP&#x2F;2r2PK1&#x2F;8 b - - 0 1) takes around 10 seconds for Stockfish.<p>In summary, chess engine might not really &quot;understand&quot; these positions but they solve them pretty well.
评论 #27191600 未加载
Andrex大约 4 年前
<a href="https:&#x2F;&#x2F;www.youtube.com&#x2F;watch?v=Z4F7mUUjt_c" rel="nofollow">https:&#x2F;&#x2F;www.youtube.com&#x2F;watch?v=Z4F7mUUjt_c</a><p>Maybe a more realistic scene than the fanbase gives it credit for. :)
评论 #27190899 未加载
TchoBeer大约 4 年前
I think there are some engines (Crystal is the one I&#x27;m thinking of) which do well in fortresses; these come at the cost of play strength.
评论 #27189287 未加载
gweinberg大约 4 年前
Has anyone actually checked that a modern chess engine believes black is winning in the Penrose position? I find it very hard to believe.
评论 #27189885 未加载
评论 #27189883 未加载
评论 #27190060 未加载
ngcc_hk大约 4 年前
Politics are about making up rules that others have to follow. May be AI still not good at playing this level of politics.
vmception大约 4 年前
Yeah I would have thought these engines to be alot smarter then this, especially the “AI” ones
johnklos大约 4 年前
Nice try, chess.com. I&#x27;m not going to let you use all those cycles stolen by your site to win chess games :P
unnouinceput大约 4 年前
Quote: &quot;Since IBM&#x27;s Deep Blue defeated World Chess Champion Garry Kasparov in their 1997 match...&quot;<p>Deep Blue lost in 1996. Its upgrade, called Deeper Blue is the one that won the famous match in 1997. Please SamCopland, do your homework.
评论 #27190701 未加载