Stockfish 14 has been released

[Cornfed]

Click on Navigation, you will see that clicking on square D8 shows the PGN so far.

Direct link - http://rebel13.nl/pgn4web/match2.html

Thanks. I do that and see all the games in a pgn txt file...how do download them as a file with all the pgn games in it escapes me however.

Using Windows, I clicked the D8 square on the chessboard. That produced an html file with all the games. I then right clicked the mouse button, clicked Select All and then Copy. Pasted the contents into Notepad and saved as sf13vsf14 rebel.pgn

Perfect - thanks~!

[dkappe]

Sorry to me it looks like the resignation, a white flag of human chess programmers who burried history of human programming dignity to cope with an enemy that uses AI.

They never understood WHY they lost to LCO.
When e.g. Andrew Grant says HCE is coming to the end…



viewtopic.php?f=2&t=77571


Its the coclusion that human designed evaluation together with search cannot compete with NN.

Its a resignation.


Or the consideration that the last years were all lost because they all relied on search instead of evaluation.

It was easier to go a ply deeper instead to do a working evaluation.

And today stockfish goes 40-50 searches deep depending on your hardware and still cannot beat lc0.

Thats a waterloo of search paradigm.

Like Andy, you are mourning your loss of relevance. Rather than learning a new skill set, you call those using new technologies names. Moving from HCE’s to those superior technologies isn’t a resignation, but a step forward.

[mclane]

For years computerchess goes into the wrong direction.
As a result of this elo increased but programs evaluation reduced. Nobody complained because ELO increased.

This came to the surface when LC0 appeared.

Suddenly NN gets “integrated”.

[mclane]

Sorry to me it looks like the resignation, a white flag of human chess programmers who burried history of human programming dignity to cope with an enemy that uses AI.

They never understood WHY they lost to LCO.
When e.g. Andrew Grant says HCE is coming to the end…



viewtopic.php?f=2&t=77571


Its the coclusion that human designed evaluation together with search cannot compete with NN.

Its a resignation.


Or the consideration that the last years were all lost because they all relied on search instead of evaluation.

It was easier to go a ply deeper instead to do a working evaluation.

And today stockfish goes 40-50 searches deep depending on your hardware and still cannot beat lc0.

Thats a waterloo of search paradigm.

Like Andy, you are mourning your loss of relevance. Rather than learning a new skill set, you call those using new technologies names. Moving from HCE’s to those superior technologies isn’t a resignation, but a step forward.

What do you learn from NN ?!

[AndrewGrant]

Sorry to me it looks like the resignation, a white flag of human chess programmers who burried history of human programming dignity to cope with an enemy that uses AI.

They never understood WHY they lost to LCO.
When e.g. Andrew Grant says HCE is coming to the end…



viewtopic.php?f=2&t=77571


Its the coclusion that human designed evaluation together with search cannot compete with NN.

Its a resignation.


Or the consideration that the last years were all lost because they all relied on search instead of evaluation.

It was easier to go a ply deeper instead to do a working evaluation.

And today stockfish goes 40-50 searches deep depending on your hardware and still cannot beat lc0.

Thats a waterloo of search paradigm.

Like Andy, you are mourning your loss of relevance. Rather than learning a new skill set, you call those using new technologies names. Moving from HCE’s to those superior technologies isn’t a resignation, but a step forward.

What do you learn from NN ?!

I too don't feel there is much to learn from NN. Extracting meaningful observations from the weights of a black box is not very tractable right now. Although I would imagine there is work on it, and I imagine again that in the future it will be common place to extract knowledge from the weights.

However, I have been learning about NN. And I was doing non-NN machine learning prior to Leela stuff. So I've certainly not strayed from the new skill set. Just complained that it takes the chess out of chess, which I believe is objectively true.

[connor_mcmonigle]

https://microscope.openai.com/models

It's possible to extract pretty meaningful information from the weights of NN using approaches such as the above. I think the insights provided by the terms of a HCE were largely overblown anyways. Evaluation functions found in top engines already contained nonlinearities which made standard techniques for extracting information from the terms of linear models impossible...

[AndrewGrant]

https://microscope.openai.com/models

It's possible to extract pretty meaningful information from the weights of NN using approaches such as the above. I think the insights provided by the terms of a HCE were largely overblown anyways. Evaluation functions found in top engines already contained nonlinearities which made standard techniques for extracting information from the terms of linear models impossible...

Hmm.. suppose my statement was not quite true then when I made it. And will continue to become less true until its patently false. Interesting.

[dkappe]

However, I have been learning about NN. And I was doing non-NN machine learning prior to Leela stuff. So I've certainly not strayed from the new skill set. Just complained that it takes the chess out of chess, which I believe is objectively true.

I included you in the mourning but not in the refusal to learn new things. Hats off.

[Rebel]

Click on Navigation, you will see that clicking on square D8 shows the PGN so far.

Direct link - http://rebel13.nl/pgn4web/match2.html

Thanks. I do that and see all the games in a pgn txt file...how do download them as a file with all the pgn games in it escapes me however.

Ah, I forgot to add a button, for the moment -

http://rebel13.nl/b/grl.pgn

[Rebel]

Code: Select all

Running      : Match SF14 vs SF13 - using Noomen openings
Time Control : 40/120
Games        : 3498

Results from file match-sf14.pgn:

No. Name          Win Draw Loss Unf.  Score Games       %
---------------------------------------------------------
  1 Stockfish 14 +414 =2914 -170   *0 1871.0  3498   53.5%
  2 Stockfish 13 +170 =2914 -414   *0 1627.0  3498   46.5%

Match finished.

+25

Games : http://rebel13.nl/b/grl.pgn