crem wrote: ↑Fri Aug 03, 2018 7:22 am
FYI, ASilverreplied to my message in LCZero forum:
Alexander,
I have pretty much said all I have to say on the matter, though I have no control over whether you accept my statements or not. I do not see any constructive outcome for any parties in pursuing a public argument on this, so I must decline this invitation. Should you have anything in private you want to talk about, you have my email.
Peace,
Albert
So, there will be no answers to those 5 questions.
Error323 wrote: ↑Thu Aug 02, 2018 4:23 pm
This was my reasoning and initial reaction also, though not public. I won't lie, when reading this it kinda comes back. This could have been done with a lot more grace. I still think it's ridiculous to call it a new engine. I can train a good network on human data in less than a week. Whatever.
In computer chess nobody cares about copyrights, licensing, plagiarism or attribution and most of the audience doesn't care either, they just want the strongest engine and that's it. Trust me on this: you can accept it or move on. You can't change the audience, there's overwhelming evidence from the past and present.
Question: if another engine were to use Leela Zero's network, is it a clone/derivative work or not? What if it's a "new" network, just trained from the same data? If I make a "new" engine with LZ's network and a bit of search optimization and a TensorRT backend, can I call it mine? I already have a good name for it: "Little Fish".
I was asked to comment on Leela Zero Chess' participation for the WCCC (my response: It's >90% Alexander's code now so he's the only one who can decide according to ICGA rules) and I raised the question whether I could enter the lczero client with the same network, but they declined to answer. It's an interesting question which will need to be resolved at some point.
In any case, what is right is right and what is wrong is wrong. Open source is not equivalent to free of copyrights.
Problem with LC0 is that their fans are on steroids and think they can speak for the authors, and once a while there may be a "new version" of their position. When the demands are not met, they RIOT! So it's kinda tricky for any other parties to follow, I suggest we should just avoid them all in order not to find ourselves in the middle of an uncivilized dogfight, WCCC wise enough to not give you an answer, yet, see what happened.
Error323 wrote: ↑Thu Aug 02, 2018 4:23 pm
This was my reasoning and initial reaction also, though not public. I won't lie, when reading this it kinda comes back. This could have been done with a lot more grace. I still think it's ridiculous to call it a new engine. I can train a good network on human data in less than a week. Whatever.
In computer chess nobody cares about copyrights, licensing, plagiarism or attribution and most of the audience doesn't care either, they just want the strongest engine and that's it. Trust me on this: you can accept it or move on. You can't change the audience, there's overwhelming evidence from the past and present.
Question: if another engine were to use Leela Zero's network, is it a clone/derivative work or not? What if it's a "new" network, just trained from the same data? If I make a "new" engine with LZ's network and a bit of search optimization and a TensorRT backend, can I call it mine? I already have a good name for it: "Little Fish".
I was asked to comment on Leela Zero Chess' participation for the WCCC (my response: It's >90% Alexander's code now so he's the only one who can decide according to ICGA rules) and I raised the question whether I could enter the lczero client with the same network, but they declined to answer. It's an interesting question which will need to be resolved at some point.
Well the short answer is: If it's legal, you can do it. That being said, I think one should apply the rule: "if I'd reverse the situation would I be ok with this?". This does require empathy and being honest with yourself.
Reversed, if someone took the codebase and managed to beat the current opensource effort he/she should deserve credit. It's a job well done, now of course it would be great if that person fed back the knowledge in the community, but this is not required.
Error323 wrote: ↑Thu Aug 02, 2018 4:23 pm
This was my reasoning and initial reaction also, though not public. I won't lie, when reading this it kinda comes back. This could have been done with a lot more grace. I still think it's ridiculous to call it a new engine. I can train a good network on human data in less than a week. Whatever.
In computer chess nobody cares about copyrights, licensing, plagiarism or attribution and most of the audience doesn't care either, they just want the strongest engine and that's it. Trust me on this: you can accept it or move on. You can't change the audience, there's overwhelming evidence from the past and present.
Question: if another engine were to use Leela Zero's network, is it a clone/derivative work or not? What if it's a "new" network, just trained from the same data? If I make a "new" engine with LZ's network and a bit of search optimization and a TensorRT backend, can I call it mine? I already have a good name for it: "Little Fish".
I was asked to comment on Leela Zero Chess' participation for the WCCC (my response: It's >90% Alexander's code now so he's the only one who can decide according to ICGA rules) and I raised the question whether I could enter the lczero client with the same network, but they declined to answer. It's an interesting question which will need to be resolved at some point.
The question is not about originality at all. I don't think people would object DeusX participating if it was done right. It's an open question when engine becomes "different", and we are not the first to have this question.
It's about ASilver's (successful) attempt submit his TCEC entry by misleading TCEC admins, and deliberate hiding the fact that he's going to use Lc0 code, from TCEC fans and LCZero devs.
I don't know if you watched the interview, but if you didn't, it's highly recommended in order to understand what's going on.
Gian-Carlo Pascutto wrote: ↑Fri Aug 03, 2018 8:50 am
In computer chess nobody cares about copyrights, licensing, plagiarism or attribution and most of the audience doesn't care either, they just want the strongest engine and that's it. Trust me on this: you can accept it or move on. You can't change the audience, there's overwhelming evidence from the past and present.
Question: if another engine were to use Leela Zero's network, is it a clone/derivative work or not? What if it's a "new" network, just trained from the same data? If I make a "new" engine with LZ's network and a bit of search optimization and a TensorRT backend, can I call it mine? I already have a good name for it: "Little Fish".
I was asked to comment on Leela Zero Chess' participation for the WCCC (my response: It's >90% Alexander's code now so he's the only one who can decide according to ICGA rules) and I raised the question whether I could enter the lczero client with the same network, but they declined to answer. It's an interesting question which will need to be resolved at some point.
I can understand where you are coming from. But maybe we can improve on how previous events were handled. There are clearly new issues to be deliberated and discussed. The goal of such a process should be a consensus of all major parties involved, for example by voting. I am not advocating for or against implementing a specific interpretation of the clone rule. What i am proposing is to give all involved the breathing space and time to do so and not get pressured by Albert Silver's covert actions . I am sure your voice will be heard in this matter, since you are the grandfather of the project.
In my opinion TCEC should reconsiders and disallows Deus X this season. This action would signal that TCEC is ready to at least discuss these issues with their main stars: the engine authors. If at the end of this process the consensus is reached that Albert Silver is ok to participate next seasno, so be it. But right now Albert Silver is pulling a fast one on the lc0 team, the engine programmers and the community to deliberately avoid this discussion. Maybe he comes to his senses and withdraws voluntarily to give the community the time to settle these matters. If so, I would welcome him back for next season with a new engine written by himself or if the consensus allows using lc0 and a custom NN.
Error323 wrote: ↑Thu Aug 02, 2018 4:23 pm
This was my reasoning and initial reaction also, though not public. I won't lie, when reading this it kinda comes back. This could have been done with a lot more grace. I still think it's ridiculous to call it a new engine. I can train a good network on human data in less than a week. Whatever.
In computer chess nobody cares about copyrights, licensing, plagiarism or attribution and most of the audience doesn't care either, they just want the strongest engine and that's it. Trust me on this: you can accept it or move on. You can't change the audience, there's overwhelming evidence from the past and present.
Fact indeed.
So then, since you knew that in advance, instead of complaining, protect your work.
Been browsing the Lc0 pages, couldn't find a clear statement about the legal status, is it freeware, GPL?
90% of coding is debugging, the other 10% is writing bugs.
Error323 wrote: ↑Fri Aug 03, 2018 9:50 am
That being said, I think one should apply the rule: "if I'd reverse the situation would I be ok with this?". This does require empathy and being honest with yourself.
I think a common failing here is a misunderstanding in what goes into writing a strong engine from scratch, and improving an existing engine by X Elo. If you have never done the former and have no idea about the work involved, it's (apparently) hard to apply that rule of thumb.
At least that would also explain why it keeps happening.
