News for hackers. Stuff that matters.

A comment on Adversarial Policies Beat Superhuman Go AIs. Parent comment

Re:Adversarial Policies Beat Superhuman Go AIs by mafuy on Thursday December 26, 2024 @10:13AM (#42515667)
This is not a proper test for chess memory. The point is that chunking is learned. A strong player has a single entity in their mind that describes a complex pattern. Magnus may have many more of these available than his opponents, letting him memorize and recall chess patterns more easily.

The same is not true for other things to memorize. "Memorizing highlighted circles on a grid" is quite different from chess patterns. So I don't think this test can be used to say anything about his relative capabilities.

[ Parent | on HN ]
Re:Adversarial Policies Beat Superhuman Go AIs by ANewFormation on Friday December 27, 2024 @08:01AM (#42521761)
Chess memory is not so rigid or limited. For instance another little demo from the same event had the players name what game a position was from, but the pieces were replaced with checker pieces. Nonetheless the players were easily able to do so.

This is why I said the test was quite adjacent to chess.

[ Parent | on HN ]