4 ms·

So I actually checked the problems listed. In the "IQ Test #52" position (FEN: 8/1p1q1k2/1Pp5/p1Pp4/P2Pp1p1/4PpPp/1N3P1P/3B2K1 w - - 0 1) listed in #1 both LC0

So I actually checked the problems listed. In the "IQ Test #52" position (FEN: 8/1p1q1k2/1Pp5/p1Pp4/P2Pp1p1/4PpPp/1N3P1P/3B2K1 w - - 0 1) listed in #1 both LC0 and Stockfish play the correct line on my computer in seconds. Both of them also play the right move Ba4+ in the second position of #1 "William Rudolph vs." (FEN: 8/1p1q1k2/1Pp5/p1Pp4/P2Pp1p1/4PpPp/1N3P1P/3B2K1 w - - 0 1) but they take quite a bit longer. Stockfish variants get it quicker. Stockfish solves "Hasek vs." (FEN: r7/7k/5R2/p3p3/Pp1pPp2/1PpP1Pp1/K1P3P1/8 w - - 0 1) listed in #2 quickly. Both Stockfish and LC0 solve "Lazard=F vs." (FEN: q7/8/2p5/B2p2pp/5pp1/2N3k1/6P1/7K w - - 0 1) quickly. Stockfish gets Bh3 from #3 "Veselin Topalov (?) vs. Alexey Shirov" (FEN: 8/8/4kpp1/3p1b2/p6P/2B5/6P1/6K1 b - - 2 47) in seconds (7-piece end game tablebases installed). The next position from #3 "Spassky, Boris V vs. Byrne, R." (FEN: 3B4/1r2p3/r2p1p2/bkp1P1p1/1p1P1PPp/p1P4P/PPB1K3/8 w - - 0 1) is easy for both Stockfish and LC0. They both get 50. c5!! right away. Stockfish also gets the last position from #3 "Stefan Brzozka vs. David Bronstein" in seconds (FEN: 1r6/4k3/r2p2p1/2pR1p1p/2P1pP1P/pPK1P1P1/P7/1B6 b - - 0 48) Rxb3+. Stockfish and LC0 see Kd1 in "Lamford=P vs." from #4 (FEN: 8/8/8/1k3p2/p1p1pPp1/PpPpP1Pp/1P1P3P/QNK2NRR w - - 0 1) but believe Rg2 is also winning. Stockfish gets c8N from "Randviir=J vs." (FEN: 5nr1/2Pp2pk/3Pp1p1/4P1P1/6P1/5K2/8/7n w - - 0 1) in #4 in about 2 minutes on my computer. Stockfish gets Bf5 from "Simkhovich=F vs." (FEN: 8/8/2pk4/8/p1p3B1/PpP5/1P6/r1NK4 w - - 2 2) in #5 in seconds. LC0 also gets it. The next two positions are mentioned as easier for programs and they are: Qe3 from "Deep Blue vs. Garry Kasparov" in #6 (FEN: 1r6/5kp1/RqQb1p1p/1p1PpP2/1Pp1B3/2P4P/6P1/5K2 b - - 14 45) is very easy for both Stockfish and LC0. Both also get "Vladimir Kramnik vs. Peter Leko" (FEN: 6k1/5p1p/P1pb1nq1/6p1/3P4/1BP2PP1/1P1Nb2P/R1B3K1 b - - 0 25) in #6 quickly. "Matous=M vs." (FEN: n2Bqk2/5p1p/Q4KP1/p7/8/8/8/8 w - - 0 1) is indeed harder for Stockfish and LC0 than expected. I've confirmed mate in 13 in my mate solver. In "Nigel Short vs. Vladimir Kramnik" (FEN: r3r1k1/1bp1Bppp/pb1p4/1p6/1P6/1BP2P2/P1P2PKP/R3R3 b - - 6 19) from #9, the engines like ...a5 more than ...c6. Hard to say that this move doesn't also win without more analysis. Stockfish wants to play c6! and b4! from "Marwitz=J vs." from #10 (FEN: 2K3k1/1p6/R3p1p1/1rB1P1P1/8/8/1Pb5/8 w - - 0 1) right from the start. LC0 takes longer but gets it as well. In "Anish Giri vs. Maxim Rodshtein" (FEN: 8/5pkp/3p1np1/Rpr5/8/6P1/PB3PKP/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. Last position "IQ Test #16" (FEN: 5k2/4bp2/2B3p1/1P4p1/3R4/3P2PP/2r2PK1/8 b - - 0 1) takes around 10 seconds for Stockfish. In summary, chess engine might not really "understand" these positions but they solve them pretty well.

mquander 5y ago
I don't think your analysis makes sense, because seeing the first move is not relevant in most of these problems. For example, in the Lamford study: https://lichess.org/analysis/8/8/8/1k3p2/p1p1pPp1/PpPpP1Pp/1P1P3P/QN1K1NRR_w_K_-_0_1 https://lichess.org/analysis/8/8/8/1k3p2/p1p1pPp1/PpPpP1Pp/1... If you spot the computer the first twelve moves and create this position: https://lichess.org/analysis/8/8/8/1k3p2/p1p1pPp1/PpPpP1Pp/1P1P3P/QNRKRN2_w_-_-_0_1#0 https://lichess.org/analysis/8/8/8/1k3p2/p1p1pPp1/PpPpP1Pp/1... My Stockfish still fails to see the key idea of Rc2 followed by Ka2 (at least for a few minutes until I got bored) -- it just wants to shuffle the rooks around forever. I'm sure that some engines given enough time will solve some of these, but your check isn't accurately assessing their capabilities.
- zone411 5y ago
  No, my Stockfish 13 definitely sees the right solution to 8/8/8/1k3p2/p1p1pPp1/PpPpP1Pp/1P1P3P/QNRKRN2 w - - 0 1 in less than 10 minutes. The score shoots up to over +20 in 6 minutes and it sees both Rc2 and Qa2. I generally looked at the winning lines, not just at the first moves. Any other positions where you think I was wrong?

mquander 5y ago

I don't think your analysis makes sense, because seeing the first move is not relevant in most of these problems. For example, in the Lamford study: https://lichess.org/analysis/8/8/8/1k3p2/p1p1pPp1/PpPpP1Pp/1P1P3P/QN1K1NRR_w_K_-_0_1 https://lichess.org/analysis/8/8/8/1k3p2/p1p1pPp1/PpPpP1Pp/1... If you spot the computer the first twelve moves and create this position: https://lichess.org/analysis/8/8/8/1k3p2/p1p1pPp1/PpPpP1Pp/1P1P3P/QNRKRN2_w_-_-_0_1#0 https://lichess.org/analysis/8/8/8/1k3p2/p1p1pPp1/PpPpP1Pp/1... My Stockfish still fails to see the key idea of Rc2 followed by Ka2 (at least for a few minutes until I got bored) -- it just wants to shuffle the rooks around forever. I'm sure that some engines given enough time will solve some of these, but your check isn't accurately assessing their capabilities.

zone411 5y ago
No, my Stockfish 13 definitely sees the right solution to 8/8/8/1k3p2/p1p1pPp1/PpPpP1Pp/1P1P3P/QNRKRN2 w - - 0 1 in less than 10 minutes. The score shoots up to over +20 in 6 minutes and it sees both Rc2 and Qa2. I generally looked at the winning lines, not just at the first moves. Any other positions where you think I was wrong?