A wise man once said "A Meijin in his twenties is inconceivable".
Although I guess Shibano will get there in a month.

Shibano's win is just another reminder that go is not an art: it is a sport for young people

_________________
Sorin - 361points.com

Move 224 illustrates that AI with too few playouts shouldn't be trusted. After a few hundred playouts, LZ157 thinks that Black leads by 80%, but after more thinking, LZ is completely confused and thinks that almost every point has about 9% winrate.

Yes, evaluations on low playouts are less reliable, particularly in positions which are not quiet, positional judgement type ones, but have long sequences which need reading out and who is winning changes sharply based on what happens many moves down the line. Ko fights, ladders and semeais being such positions bots need extra playouts to judge well.


Doesn't look confused to me: looks like it now correctly understands black is hopeless and every move is bad.

Here's the game with KataGo analysis (but Lizzie 0.7 puts the output in a new LZ sgf tag instead of comment so eidogo player doesn't show it): it had trouble understanding the ko at the end (thinking black was winning): failing to see that white had enough threats if he played point-losing ones which allowed him to win it.

Here is Uberdude's file with the LZ tag replaced with C.

_________________
Dave Sigaty
"Short-lived are both the praiser and the praised, and rememberer and the remembered..."
- Marcus Aurelius; Meditations, VIII 21