Call to all Engine Programers to complain to TCEC

[IQ]

To all Programmers, Participants of TCEC, TCEC leadership and computer chess afficionados,

We can still stop con-artist and self proclaimed ai-guru Albert Silver from getting away with his biggest coup. Please help the lc0 team that has been ambushed and taken advantage of. See this thread http://talkchess.com/forum3/viewtopic.php?f=2&t=68094 for details, a link to Albert Silvers "My True Colors" video, and the opinion of some of the main lc0 developers. Albert Silvers secrecy, his last minute reveal, behind the back communication with TCEC, his insistance that his TCEC participation is a done deal and immutable, did not allow the lc0 team to properly gather their thoughts and take appropriate actions. Even with the added fineprint on chessdom, only now the full scope and intent of his actions starts to sink in.

It is not to late, contrary to what Albert Silver wants everybody to believe by playing the sweet-talking, sole developer and misunderstood AI-genius. His actions are not only unfair, disrespectful to the lc0 team but also against the spirit of TCEC. Don't let him pull ai wool over your eyes by claiming that only his NN is an achievement and that search, interface code, move encoding, uci parser, training tools, etc. are just mere trivial nuisances that hinder Albert Silvers AI genius.

You as programmers should know better and not get lost in his AI smokescreen, even if not to help the lc0 team but also in self-interest to ensure a fair competition. If enough of you complain we can stop this and give the lc0 team some time to contemplate on how to handle such cases in the future, also TCEC could start a process with all of you involved that as an outcome will have a ruleset that has been discussed and agreed upon by the stars of their show: the programmers, dev teams, and collaborators of the engines participating. Do not let Albert Silver pull a fast one on everybody - which is clearly his intention! He wants to generate facts before anybody has the time to think.

Marco, Gary and all the others involved in SF development. Please help!

Larry and Mark: Please help!

Ubaldo, Norman, Robert, Daniel, Youri, Frank and all the others help!

All computer chess afficionados, please help!

If you complain to TCEC we can stop this, even if TCEC has already started. I am sure TCEC will listen. This would allow us all to take a deep breath, discuss the ramifications of ongoing and upcoming issues like this and come to a community consensus, without being pressured by Albert Silvers pulling a fast one on us and in a way that respects the lc0 dev team. Also the lc0 devs that have not yet spoken up in the other thread are encouraged to do so here!

[Graham Banks]

What TCEC does is their own business.

[noobpwnftw]

No thanks, I find it more interesting as is.
The authors of LC0 gets to decide their stand on the matter of how their software may be used, your attempt to riot on TCEC seems low.

[corres]

To all Programmers, Participants of TCEC, TCEC leadership and computer chess afficionados,

We can still stop con-artist and self proclaimed ai-guru Albert Silver from getting away with his biggest coup. Please help the lc0 team that has been ambushed and taken advantage of. See this thread http://talkchess.com/forum3/viewtopic.php?f=2&t=68094 for details, a link to Albert Silvers "My True Colors" video, and the opinion of some of the main lc0 developers. Albert Silvers secrecy, his last minute reveal, behind the back communication with TCEC, his insistance that his TCEC participation is a done deal and immutable, did not allow the lc0 team to properly gather their thoughts and take appropriate actions. Even with the added fineprint on chessdom, only now the full scope and intent of his actions starts to sink in.

It is not to late, contrary to what Albert Silver wants everybody to believe by playing the sweet-talking, sole developer and misunderstood AI-genius. His actions are not only unfair, disrespectful to the lc0 team but also against the spirit of TCEC. Don't let him pull ai wool over your eyes by claiming that only his NN is an achievement and that search, interface code, move encoding, uci parser, training tools, etc. are just mere trivial nuisances that hinder Albert Silvers AI genius.

You as programmers should know better and not get lost in his AI smokescreen, even if not to help the lc0 team but also in self-interest to ensure a fair competition. If enough of you complain we can stop this and give the lc0 team some time to contemplate on how to handle such cases in the future, also TCEC could start a process with all of you involved that as an outcome will have a ruleset that has been discussed and agreed upon by the stars of their show: the programmers, dev teams, and collaborators of the engines participating. Do not let Albert Silver pull a fast one on everybody - which is clearly his intention! He wants to generate facts before anybody has the time to think.

Marco, Gary and all the others involved in SF development. Please help!

Larry and Mark: Please help!

Ubaldo, Norman, Robert, Daniel, Youri, Frank and all the others help!

All computer chess afficionados, please help!

If you complain to TCEC we can stop this, even if TCEC has already started. I am sure TCEC will listen. This would allow us all to take a deep breath, discuss the ramifications of ongoing and upcoming issues like this and come to a community consensus, without being pressured by Albert Silvers pulling a fast one on us and in a way that respects the lc0 dev team. Also the lc0 devs that have not yet spoken up in the other thread are encouraged to do so here!

It is pity, but if the developers work on an open source product they result can be used by any others.
You can call the users to account the concerning rules only.

[Gian-Carlo Pascutto]

I don't think TCEC ever had real rules regarding program originality (that don't amount to: whatever the organizer judges to be an original enough engine is one) so I don't see why you're expecting them to suddenly care.

[Laskos]

So much rabidness. As long as there are only two Leela's (lc0) in TCEC, one original, another a clone, one using reinforcement learning, another supervised learning, I wouldn't be very bothered. Albert's work is still Leela to me. But I don't have 2x1080ti and LTC to see how these two networks behave. If supervised learning comes out as performing well, I might fiddle myself around with Leela networks (never claiming anything as my own aside the network). The problem would be legality of what Albert does, especially when commercial issues are concerned. Also, next time, maybe 5 Leela clones will appear, and that's a problem. By the way, the current testent is progressing very rapidly against AB engines (not solely in their selfgames), it's already about 3300+ CCRL 40/4' Elo level on GTX 1060. The testnets are the strongest nets overall (stronger than mainserver nets), and scaling very well to LTC.

Just my small thoughts, I am not an author of anything, so my opinions here have little relevance.

[IQ]

So much rabidness. As long as there are only two Leela's (lc0) in TCEC, one original, another a clone, one using reinforcement learning, another supervised learning, I wouldn't be very bothered. Albert's work is still Leela to me. But I don't have 2x1080ti and LTC to see how these two networks behave. If supervised learning comes out as performing well, I might fiddle myself around with Leela networks (never claiming anything as my own aside the network). The problem would be legality of what Albert does, especially when commercial issues are concerned. Also, next time, maybe 5 Leela clones will appear, and that's a problem. By the way, the current testent is progressing very rapidly against AB engines (not solely in their selfgames), it's already about 3300+ CCRL 40/4' Elo level on GTX 1060. The testnets are the strongest nets overall (stronger than mainserver nets), and scaling very well to LTC.

Just my small thoughts, I am not an author of anything, so my opinions here have little relevance.

I can fully understand your curiosity in how different NN would perform at TCEC. We are all impatient regarding advances in a field dear to our heart. On the other hand postponing the urge to satisfy one's curiosity a couple of month until next season seems a small price to pay when considering how many open issues remain. This is not even taking into consideration the disrespectful treatment of the lc0 dev team. Wouldn't you rather see original engines that apart from being NN based also include other innovative ideas? How interesting would it be to see an army of leelas playing with endless variations of NN? How fair would that be to the competition? Where does this leave SF clones with changed eval? SF is also open source, yet clones are not allowed? And do we really want to encourage a mentality that rewards those that boldly overstep ethical bounds without considering any of the other parties involved? I think we have to weigh that against our short term curiosity.

[noobpwnftw]

If I have to be honest to my heart, this isn't about the legitimacy or anything, I guess the LC0 authors are at least cool with the current clarifications, but the real storm is yet to come.

Fanboys' worst nightmare is that his NN may actually perform better in practice, which instead of learning from zero and a massive collection of resources, even or better performance may be achieved with supervised training on human games from a one-man's job, while they share the same engine, that's the big OUCH.

Now that is what this is all about, no? If that NN isn't anywhere close to challenge the "zero superiority" status, then all this drama can never be so heated and people probably would just laugh at that guy and move on.

[elcabesa]

I haven't followed Deus X case very closely, but those are my thoughts.
1) major leela authors, who are the owner of the copyright, if they feel Deus is against rules, or against copyright shall formally complain.
2) from my point of view zero and supervised approach are interesting to be evaluated against in a match.
3) Tcec is taking big risk with those decision, because it will create strange precedent for next seasons.

Marco

[whereagles]

pseudo-code for alpha-beta is known for a long time.. any implementation is bound to be extremely similar to a previous one, if not equal.

if this isn't a problem, why should using two different NN be?