Human Beats Top Go Computer

Discussion of anything and everything relating to chess playing software and machines.

Moderator: Ras

User avatar
towforce
Posts: 12682
Joined: Thu Mar 09, 2006 12:57 am
Location: Birmingham UK
Full name: Graham Laight

Re: Human Beats Top Go Computer

Post by towforce »

More evidence in support of of today's artificial NNs being more like a cage of parrots than a deep, sophisticated model of their target knowledge area: a recent paper asserts that emergent behaviours in LLMs (large language models) are a mirage. To be clear, they haven't proven that emergent behaviours don't exist, but they have shown that previous work that seems to show that they do is flawed because is was measuring discrete "all or nothing" steps, not a continuous scale.

https://www.theregister.com/2023/05/16/ ... _behavior/
Human chess is partly about tactics and strategy, but mostly about memory
User avatar
towforce
Posts: 12682
Joined: Thu Mar 09, 2006 12:57 am
Location: Birmingham UK
Full name: Graham Laight

Re: Human Beats Top Go Computer

Post by towforce »

Improved Analogy


"Pile of flashcards" is a better analogy than "cage of parrots".

I'm crediting a video about learning I've just watched (link) which states that flashcards are good for learning snippets of information, but not good for learning concepts. This nicely summarises my view about today's artificial NNs!
Human chess is partly about tactics and strategy, but mostly about memory
User avatar
towforce
Posts: 12682
Joined: Thu Mar 09, 2006 12:57 am
Location: Birmingham UK
Full name: Graham Laight

Re: Human Beats Top Go Computer

Post by towforce »

I'm not the only person to realise that big, knowledgeable NNs are like parrots - but, apparently, they're "stochastic parrots", not "a cage of parrots": https://nymag.com/intelligencer/article ... ender.html (a long article, but worth reading).

I still prefer "pack of flashcards" (or "deck of flashcards" if you're American): for me, it expresses what I've been trying to say really well.
Human chess is partly about tactics and strategy, but mostly about memory
Chessqueen
Posts: 5685
Joined: Wed Sep 05, 2018 2:16 am
Location: Moving
Full name: Jorge Picado

Re: Human Beats Top Go Computer

Post by Chessqueen »

towforce wrote: Thu May 04, 2023 9:36 pm Kellin Pelrine, an amateur Go player, and his team uncovered weaknesses in the play of the top Go computers - a mistake a human simply wouldn't make (misunderstanding the nature of a group of stones). They built a program that exploited this weakness, and it consistently beat the top Go program.
No BIG deal, I constantly beat all of these Chess Engines listed here, whenever I want to :roll: :mrgreen: :roll:

526 Cicada 0.1 64-bit 1538 +24 −25 44.1% +48.0 19.7% 636
65.6%
527 MSCP 1.6h 1531 +25 −25 44.0% +50.4 15.1% 628
99.8%
528 Apep 0.1.0 1480 +25 −25 41.2% +72.8 10.1% 683
79.1%
529 Saruman 2017.08.10 64-bit 1468 +18 −18 30.7% +151.4 27.2% 1249
95.8%
530 Pooky 2.7 1441 +26 −26 31.5% +144.3 17.1% 626
75.7%
531 Dreamer 0.3.0 64-bit 1427 +25 −26 33.9% +122.5 16.9% 640
96.5%
532 Hokus-Pokus 0.6.3 1391 +28 −16 26.7% +197.2 11.0% 628
72.6%
533 BACE 0.46 1380 +25 −6 21.5% +256.6 9.8% 1027