| Life In 19x19 http://lifein19x19.com/ |
|
| O Meien on how to study with AI bots http://lifein19x19.com/viewtopic.php?f=13&t=16108 |
Page 1 of 1 |
| Author: | John Fairbairn [ Sun Sep 30, 2018 4:49 am ] |
| Post subject: | O Meien on how to study with AI bots |
O Meien has published the last in his long series of "AlphaGo Reports" in Go World this month, and the article is a combination of his conclusions and tips for studying with AI. His conclusions are rather banal (go can now be divided into pre-AG and post-AG eras), and evidently he regards AlphaGo Zero as the pinnacle but is no nearer working out the mysteries of what it is doing in the middle game. He also concludes that go still has appeal post-AG, though the very fact he has to state that perhaps indicates he's had to think very hard about it. The most interesting thing he said was he uses Leela Zero to study his own games. He didn't say what setup he uses but just pointed readers at the open source (and FWIW with a low-grade setup I got the same sort of results he did). He seems to regard it as stronger than himself. He gave an example. This position is from a very recent game of his. He was White and had just played the triangled stone, which his opponent answered at A. O was then a bit beflummoxed because it seemed natural for him to continue with the ponnuki at B. He didn't know it then, but LZ doesn't even consider this move, and indeed O had second thoughts when he realised that after the capture at B (and Black C) there are no moves he really wants to play as White. LZ homes in on a single move, C, and O found that move for himself. It's still not very good for him (and he lost the game) but he was puzzled about his triangled move. He therefore stepped back and looked at what LZ preferred and was surprised to see that LZ chose D, which in fact gives White a reasonable game. He doesn't explain why he was surprised, but there he gives strong hints that was trying to avoid life and death hassles in the corner and that he was also thinking about forcing moves - presumably he meant he couldn't see the wood for the trees. Anyway, the upshot seems to be that we amateurs keeping tabs on LZ are plodging though the same muddy roads as the pros. |
|
| Author: | Uberdude [ Sun Sep 30, 2018 5:48 am ] |
| Post subject: | Re: O Meien on how to study with AI bots |
I'm struggling a bit to work out a move order in which O would play the triangled atari and not expect black to munch the corner (ko in tripod group with extra leg notwithstanding). Maybe like this: Big extend to 8 is in fashion in post-AG world. White has ladder which probably explains black's choice of 17 not 19 (but given that why not 15 as extend not hane? LZ agrees its better). 20 invasion is follow-up of ignored 18. 23 attach is a fairly standard tesuji in the shape which helps deal with the cut at e should the standard corner sequence a-d happen. Not sure when black would have played it but this order seems plausible, Then O's atari of 28 instead of slide to a is presumably fighting spirit and trying to make black's attachment into a bad exchange, but focusing too much on that and not how his 3-3 invasion stones also became bad exchanges. Reminds me rather of Lee Sedol vs AG game 3 at lower left corner. |
|
| Page 1 of 1 | All times are UTC - 8 hours [ DST ] |
| Powered by phpBB © 2000, 2002, 2005, 2007 phpBB Group http://www.phpbb.com/ |
|