Surprise: Official release version of Stockfish 14.1

[Jouni]

This come out of nowhere!

[bmp1974]

I feel there was nothing wrong in releasing it as Stockfish 15. After all, 17 elo gain over SF 14 is not a small thing at this level.

[marsell]

This way, the clones do not get the opportunity to become number one in rankings (see Fat Fritz).

[Andrew]

I hope they do a new net version soon, the last one on the abrok site was 15 September.

The other thing they could consider is having a version with a larger net size than the main version.
Might be useful for longer TC games or long analysis?

Andrew

[Deberger]

This come out of nowhere!

Indeed. Especially as some recent changes show negative Elo !

https://github.com/glinscott/fishtest/w ... sion-Tests

I hope they do a new net version soon, the last one on the abrok site was 15 September.

Indeed. But no nets are attempted in the past 12 days.

https://tests.stockfishchess.org/nns

[Cornfed]

This come out of nowhere!

Indeed. Especially as some recent changes show negative Elo !

https://github.com/glinscott/fishtest/w ... sion-Tests

The first...is not unusual. Do not dismay. Do not that the Elo+.03 or Elo+1.29, etc, etc for each development version at face value....or SF would probably be over 4000 elo by now.

[dkappe]

This come out of nowhere!

Indeed. Especially as some recent changes show negative Elo !

https://github.com/glinscott/fishtest/w ... sion-Tests

I hope they do a new net version soon, the last one on the abrok site was 15 September.

Indeed. But no nets are attempted in the past 12 days.

https://tests.stockfishchess.org/nns

Stockfish has come to rely on Leela data and has abandoned RL entirely. That means they’re converting and tweaking data for SL runs. I suspect the next major net improvement will come from a new net size or architecture.

For those of us still doing RL with adversarial training, I welcome the release of 14.1 with the traditional greeting: next victim!!

[Paloma]

What the heck is RL ?

[CMCanavessi]

Reinforced Learning?

[dkappe]

What the heck is RL ?

Two abbreviations important to game neural network training: SL = supervised learning, RL = reinforcement learning. What Leela, AlphaZero and Dragon were/are doing is RL, so the training data from the improving nets is feeding back into the training. Stockfish is currently doing SL, so just a big batch of data from leela that they try to massage into a better net.

Note that SF started out with RL which produced their initial strong nets.