Stockfish 14 release round the corner

[ChickenLogic]

Wow, this thread just made my eyes bleed. People test Stockfish clones (Houdini, Fire and other clones) without any bad feelings because it's 'just copying Stockfish'. But Stockfish mustn't EVER use anything somebody else has created or it is immediately a big scandal and 'testers' all of the sudden have moral issues. Y'all need to come back to reality. How is it ok to test Fire, which literally takes a net created only by SF for SF, but not test SF once it starts using Leela data? It's also using SF data, a trainer made by SF, own testing framework, completely unique net topology but somehow all these are not enough to compensate for taking Leela data...

Conclusion: talkchess chatters think that literally stealing a net made by and for SF while also being unable to spell the name of the creator is somehow morally acceptable but creating something completely new and taking the liberty to take freely available data isn't.

Garbage. Complete and utter garbage.

[AndrewGrant]

Conclusion: talkchess chatters think that literally stealing a net made by and for SF while also being unable to spell the name of the creator is somehow morally acceptable but creating something completely new and taking the liberty to take freely available data isn't.

Nail. Hammer. Head.

[Madeleine Birchfield]

Wow, this thread just made my eyes bleed. People test Stockfish clones (Houdini, Fire and other clones) without any bad feelings because it's 'just copying Stockfish'. But Stockfish mustn't EVER use anything somebody else has created or it is immediately a big scandal and 'testers' all of the sudden have moral issues. Y'all need to come back to reality. How is it ok to test Fire, which literally takes a net created only by SF for SF, but not test SF once it starts using Leela data? It's also using SF data, a trainer made by SF, own testing framework, completely unique net topology but somehow all these are not enough to compensate for taking Leela data...

They don't even test the Leela versions from where the Stockfish team got the data, so for all points and purposes Stockfish would be the first engine on the ratings list to use the data.

[Terje]

Wow, this thread just made my eyes bleed. People test Stockfish clones (Houdini, Fire and other clones) without any bad feelings because it's 'just copying Stockfish'. But Stockfish mustn't EVER use anything somebody else has created or it is immediately a big scandal and 'testers' all of the sudden have moral issues. Y'all need to come back to reality. How is it ok to test Fire, which literally takes a net created only by SF for SF, but not test SF once it starts using Leela data? It's also using SF data, a trainer made by SF, own testing framework, completely unique net topology but somehow all these are not enough to compensate for taking Leela data...

Conclusion: talkchess chatters think that literally stealing a net made by and for SF while also being unable to spell the name of the creator is somehow morally acceptable but creating something completely new and taking the liberty to take freely available data isn't.

Garbage. Complete and utter garbage.

This. Anyone implying the new net made with partly leela data makes it a problem to test Stockfish while having clones feature on the list needs to reevaluate what they think makes an engine 'original'.

[AndrewGrant]

Wow, this thread just made my eyes bleed. People test Stockfish clones (Houdini, Fire and other clones) without any bad feelings because it's 'just copying Stockfish'. But Stockfish mustn't EVER use anything somebody else has created or it is immediately a big scandal and 'testers' all of the sudden have moral issues. Y'all need to come back to reality. How is it ok to test Fire, which literally takes a net created only by SF for SF, but not test SF once it starts using Leela data? It's also using SF data, a trainer made by SF, own testing framework, completely unique net topology but somehow all these are not enough to compensate for taking Leela data...

Conclusion: talkchess chatters think that literally stealing a net made by and for SF while also being unable to spell the name of the creator is somehow morally acceptable but creating something completely new and taking the liberty to take freely available data isn't.

Garbage. Complete and utter garbage.

This. Anyone implying the new net made with partly leela data makes it a problem to test Stockfish while having clones feature on the list needs to reevaluate what they think makes an engine 'original'.

Conversation-ender. I'm a purist -- I want all original anywhere, but to complain about Stockfish from any other stance is laughable.

[connor_mcmonigle]

Wow, this thread just made my eyes bleed. People test Stockfish clones (Houdini, Fire and other clones) without any bad feelings because it's 'just copying Stockfish'. But Stockfish mustn't EVER use anything somebody else has created or it is immediately a big scandal and 'testers' all of the sudden have moral issues. Y'all need to come back to reality. How is it ok to test Fire, which literally takes a net created only by SF for SF, but not test SF once it starts using Leela data? It's also using SF data, a trainer made by SF, own testing framework, completely unique net topology but somehow all these are not enough to compensate for taking Leela data...

Conclusion: talkchess chatters think that literally stealing a net made by and for SF while also being unable to spell the name of the creator is somehow morally acceptable but creating something completely new and taking the liberty to take freely available data isn't.

Garbage. Complete and utter garbage.

To their credit, CCRL at least has not tested Fire with the Stockfish network. I think the context Graham is missing is that Stockfish with a network partially trained on data from the Lc0 project still takes far less from Lc0 than the likes of Igel, Rubi , Nemo, etc. take from Stockfish. Furthermore, if CCRL is going to list Stockfish clones and even list Stockfish (FF2) with a different network (trained on Lc0 data) as a separate engine, then not listing Stockfish as its latest network is partially trained on Lc0 data is just utterly ridiculous.

Fat Fritz 2: Stockfish with a network trained (entirely) on Lc0 (Fat Fritz 1) games

Oh well. Perhaps it's time for a a new rating list which is more supportive of actual chess engine developers.

[Madeleine Birchfield]

I think the context Graham is missing is that Stockfish with a network partially trained on data from the Lc0 project still takes far less from Lc0 than the likes of Igel, Rubi , Nemo, etc. take from Stockfish.

The precedent for Stockfish 14, which partially used Leela data in training, is Nemorino 6.00, which partially used Stockfish data in training, and which was tested on every ratings list including CCRL.

[dkappe]

To their credit, CCRL at least has not tested Fire with the Stockfish network. I think the context Graham is missing is that Stockfish with a network partially trained on data from the Lc0 project still takes far less from Lc0 than the likes of Igel, Rubi , Nemo, etc. take from Stockfish. Furthermore, if CCRL is going to list Stockfish clones and even list Stockfish (FF2) with a different network (trained on Lc0 data) as a separate engine, then not listing Stockfish as its latest network is partially trained on Lc0 data is just utterly ridiculous.

Fat Fritz 2: Stockfish with a network trained (entirely) on Lc0 (Fat Fritz 1) games

Oh well. Perhaps it's time for a a new rating list which is more supportive of actual chess engine developers.

Just to correct you: Fat Fritz 2 was not trained on lc0 training data.

Night Nurse was also not trained on training data (or lc0, for that matter). I’m happy to welcome SF14 to the mcts/nn data club.

[AndrewGrant]

Just to correct you: Fat Fritz 2 was not trained on lc0 training data.

I don't think you can assert that. And I would sooner assert the opposite, given Albert's inability to do anything on his own.

[dkappe]

Just to correct you: Fat Fritz 2 was not trained on lc0 training data.

I don't think you can assert that. And I would sooner assert the opposite, given Albert's inability to do anything on his own.

I can and I did. Best to sit down now and not speak of things about which you know nothing.