https://github.com/gcp/leela-zero/issues/1220
我在一个月前被两位来自中国的博士生问到,他们是否可以参加和操作Leela Zero参加本次比赛。我犹豫了一下,因为恕我直言,进入Leela Zero参加计算机比赛的机器人比较有点离题,并且减少了项目的目标
对我而言,这毫无意义,因为你可以从字面上理解权重和数据(甚至是代码,取决于道德和比赛规则),改进一点(我建议增加一些去知识!),并且你应该有一个更强大的程序你了解如何正确测试。
开源机器人和竞赛的问题并不新鲜。电脑国际象棋的历史尤其糟糕,这也是我进入编程领域的原因之一。最后,比赛组织者应该制定规则,确保公平竞争,并且实际上拥有执行这些规则的球。
这些与开源(或公开)机器人和秘密机器人的私人匹配现象并不是我以前见过的。这可能表明你现在可以更容易地隐藏在线服务背后的程序,或者只是超级公司发现可用于宣传的游戏(咳嗽IBM咳嗽Google)。用户可以购买的程序不能真正“隐藏”它的实力,因为用户可以(并且确实是CCRL,TCEC等)与其一起运行锦标赛。
另一方面,我认为不进入也不会做任何事情,是吗?通过进入,我们是众所周知的公众和其他球队击败的基准。我宁愿让更多的人了解这个项目,所有的事情都考虑在内。所以我允许进入Leela Zero。我还没有听到,但我从上面的假设,至少它进入了某种程度。
无论如何,从上述观点来看,如果有的话,即使看起来它偶尔允许其他团队得分“胜利”或PR,也不值得在程序,代码和数据的这些用途方面过于努力你认为有问题。合作努力的特点是它可以超越秘密或专有项目,并最终为用户带来更多价值。
我不知道你的情况,但我仍然使用干鱼来分析我的国际象棋比赛,而不是Alpha Zero或Deep Blue。
I got asked about a month ago by two PhD students from China whether they could enter and operate Leela Zero for this tournament. I hesitated for a bit, because IMHO entering Leela Zero into competitions for computer go bots is somewhat besides the point and detracts from the project's goal.
To me it feels pointless because you can literally take the weights and data (and even the code, depending on morals and tournament rules), improve it a bit (I suggest adding some go knowledge!), and you should have a stronger program if you understand how to test properly.
The problem of open source bots and competitions is not new. Computer chess has a particularly bad history with this and it's one of the reasons I went into go programming. In the end, it is up to the tournament organizers to set rules that ensure fair play and actually have the balls to enforce them.
The phenomenon of these private matches with open source (or just public) bots and secret bots is not something I've seen before though. It may be a sign of the times when you can now "hide" the program easier behind online services, or simply super-corporations finding games useful for publicity (cough IBM cough Google). A program that the user can buy commercially can't really "hide" it's real strength for long, as the user can (and does, CCRL, TCEC, etc) run tournaments with it.
On the other hand, I considered that not entering doesn't do anything either, does it? By entering, we are a well known benchmark for the public and other teams to beat. I'd rather have more people finding out about the project, all things considered. So I gave permission to enter Leela Zero. I have not heard back, but I assume from the above at least it got entered somehow.
Anyway, my point, if anything, from all the above is that it is not worth getting too worked up over these kind of uses of the program, code and data even if it seems it occasionally allows other teams to score "victories" or PR you consider questionable. A collaborative effort has the property that it can outlast secret or proprietary projects and eventually bring more value to the users.
I don't know about you, but I'm still using Stockfish to analyze my chess games, and not Alpha Zero or Deep Blue.
|