It is currently Thu Dec 05, 2019 2:42 pm

All times are UTC - 8 hours [ DST ]




Post new topic Reply to topic  [ 720 posts ]  Go to page Previous  1 ... 26, 27, 28, 29, 30, 31, 32 ... 36  Next
Author Message
Offline
 Post subject: Re: “Decision: case of using computer assistance in League A
Post #561 Posted: Sun Jun 17, 2018 11:03 am 
Dies in gote

Posts: 63
Liked others: 0
Was liked: 12
Rank: FFG 5k
Look at the game is useless for me: i'm 2k,so i can't judge the difference betwen a 4d-6d-bot lvl.

What i dont understand is why it would be a very difficult task for a malicious person to produce such a statement?(Np-problem)
All he have to do is to check the Carlo's moves where Leela 0.11 and Leela Zero disagree? check the moves "Leela 11 says it is very bad move but Leela Zero says it is super so he plays it", i dont know how many of them exists,and on which metrics it is based(top 3 choice,Winrate variation) but it doesn't matter,we are clearly not in a situation "impossible to find,easy to check" like PvsNp problem.

Top
 Profile  
 
Offline
 Post subject: Re: “Decision: case of using computer assistance in League A
Post #562 Posted: Sun Jun 17, 2018 12:12 pm 
Lives in sente

Posts: 1259
Liked others: 102
Was liked: 265
This post from reddit strikes me as a crock of shite.
Yes, CM could have cheated, but then why tell people about it. Is the allegation that the entire Italian squad was complicit, if so then doesn't that strike you as slightly beyond belief.
Yes okay, Leela 11 and Leela Zero differences exist, but so what? Why would CM want to cheat with the obselete Leela 11 in the first place? Why not pick 1 of many strong LZ networks?
CM certainly played very strongly this game, some proper computer analysis would be welcome, taking into account timings.

_________________
North Lecale


This post by Javaness2 was liked by: Charlie
Top
 Profile  
 
Offline
 Post subject: Re: “Decision: case of using computer assistance in League A
Post #563 Posted: Sun Jun 17, 2018 12:26 pm 
Honinbo

Posts: 9135
Liked others: 2794
Was liked: 3093
Simba wrote:
Thanks, Bill. I'm actually already fairly familiar with Regan's work having browsed chess forums in the past.

Forgive me - I'm not too eager to trawl through 29 pages of content again to check - do you recall by any chance whether or not someone has already applied Regan's methodology to any of Carlo's games?


As far as I know, no-one has come close to doing so.

First, Regan distinguishes three kinds of evidence, physical, behavioral, and statistical. Without physical or behavioral evidence he requires extremely good statistical evidence, equivalent to a difference of 5 standard deviations.

Second, go does not yet possess the tools that chess has to do the job, statistically. We are unable to reliably measure the difference in a player's level of play in a single game or group of even games and his current level of play in games with presumably no cheating. We are unable to reliably assign a difficulty rating to individual plays. We are unable to reliably order the choice of individual plays. (Yes, top bots do evaluate plays, but they do not require accurate evaluations to beat humans or the previous version of themselves.) Ten years from now, we may have good enough tools, especially if we try to develop them, but not yet.

Now we are able, to some extent, to rate the difficulty of individual plays. goproblems.com does rate the difficulty of individual problems, so in theory we could rate individual plays themselves, using only human judgement. Furthermore, players may be able to judge the difficulty of plays. Such judgements are of course imperfect, but we can gain reliability by combining the ratings of individual players. Of the 50 plays examined in the Metta-Ben David game, I judged that only 12 were difficult enough to be relevant to the question of cheating. Using similar but different criteria Bojanic found 6 important plays by Metta between move 50 and 105, a smaller range of 28 plays by him. He and I agreed on 4 of those 6 plays. My guess is that with even several judges we would have consensus on at least 3 plays. Uberdude has also tried his hand at evaluating the difficulty of individual plays. :)

Third, I think that the choice of plays itself offers behavioral evidence, not of a statistical nature, but relying upon go analysis. Stanslaw Frejlak made a go analysis of the Metta - Ben David game. (See viewtopic.php?p=228924#p228924 ). Also, in your game vs. Metta, if Black 157 descends, White can win the resulting semeai. The play is pretty much a one lane road, so is within the capabilities of a European 4 dan. OTOH, we know that in the heat of battle mistakes are made. Such go analysis is like expert testimony in court, not infallible, but not worthless, either. :)

_________________
The Adkins Principle:

At some point, doesn't thinking have to go on?

— Winona Adkins

"Once in a very great while his eyes light up for a moment, and he says "Whee!" very quietly."
— Lion Miller

Top
 Profile  
 
Offline
 Post subject: Re: “Decision: case of using computer assistance in League A
Post #564 Posted: Sun Jun 17, 2018 12:29 pm 
Honinbo

Posts: 9135
Liked others: 2794
Was liked: 3093
Simba wrote:
As it happens, the move in question (156) was Leela 0.11's third choice, according to https://www.reddit.com/r/baduk/comments ... d/e0c309j/. The ranking Leela 0.11 gives to the move is irrelevant for this discussion.


Precisely. As is its evaluation of the play. The anonymous accuser has created a diversion.

_________________
The Adkins Principle:

At some point, doesn't thinking have to go on?

— Winona Adkins

"Once in a very great while his eyes light up for a moment, and he says "Whee!" very quietly."
— Lion Miller

Top
 Profile  
 
Offline
 Post subject: Re: “Decision: case of using computer assistance in League A
Post #565 Posted: Sun Jun 17, 2018 12:32 pm 
Honinbo

Posts: 9135
Liked others: 2794
Was liked: 3093
Fenring wrote:
The very difficult problem here is to create a situation where the best move of Leela Zero is a game over for Leela 0.11.


Why bother? Let's not get distracted by shiny baubles.

_________________
The Adkins Principle:

At some point, doesn't thinking have to go on?

— Winona Adkins

"Once in a very great while his eyes light up for a moment, and he says "Whee!" very quietly."
— Lion Miller

Top
 Profile  
 
Offline
 Post subject: Re: “Decision: case of using computer assistance in League A
Post #566 Posted: Sun Jun 17, 2018 12:48 pm 
Dies in gote

Posts: 45
Liked others: 31
Was liked: 18
Rank: 2kyu
Since people have been free with their opinions, to the extent of calling others liars, etc, then I will be free with mine.

1. The organizers of the online tournament made a mistake when they reversed the Metta decision. They were faced with 3 choices, uphold the decision, reverse it or cancel the tournament. Given the situation I feel that cancelling the tournament would have been the correct choice.

2. It is a mistake for someone facing such serious allegations of cheating to work as a referee in a major tournament. It is not likely to lead to anything good.

3. Any adjudication in a tournament should be done independently and impartially and free from any meddling and or pressure from the organizers of other tournaments.

4. No cash prizes or ratings should be awarded for performance in online tournaments.

5. This should not be about any individual person, regardless of what he or she may have done. This issue is much bigger than that.

Top
 Profile  
 
Offline
 Post subject: Re: “Decision: case of using computer assistance in League A
Post #567 Posted: Sun Jun 17, 2018 1:26 pm 
Beginner

Posts: 17
Liked others: 1
Was liked: 7
Rank: NL 2 dan
KGS: MrOoijer
Simba wrote:
As it happens, the move in question (156) was Leela 0.11's third choice, ...


What does that show? Nothing IMO.

For the 5 alternatives White has for this move (h19, J-19-J16) 3 win the game if my analysis is correct. J18 played by Metta is a well known shape move. It is not even too difficult to see that it works when you notice the dame zumari of Blacks bigger dragon.

You are a 3 dan who lost to a 4 dan. What's wrong with that? Why put up a show how you let your team down etc. ? Your only "evidence" that white cheated is an unnamed person that made some remarks on reddit.

We are in phase 3 of a classical witch hunt. In phase 1 a very biased analysis seems to "prove" that the victim is guilty. Then in phase two that biased analysis is debunked by an expert, but the original hunters don't really study that but just go ahead and produce new "evidence". That is called a Gish Gallop, it is the "yes but" phase.

When that doesnt work anymore we get the "I have heard from a reliable source" phase - and that is phase 3 in which this witch hunt is now. The phase of disinformation, propaganda and fake news.

So I have a question for Ales Cieply who wrote the original "98%" analysis. That was debunked in much detail by Metta's PhD advisor Francesco Morandin, and I have never seen a detailed reply from Cieply to that report from 23 April 2018. So I have a simple qusetion: where is your reply?


This post by Jan.van.Rongen was liked by 2 people: Charlie, zermelo
Top
 Profile  
 
Offline
 Post subject: Re: “Decision: case of using computer assistance in League A
Post #568 Posted: Sun Jun 17, 2018 2:11 pm 
Lives with ko

Posts: 141
Liked others: 26
Was liked: 89
Rank: 5 dan
There is one important point that we have not discussed in this topic.
That is how big difference is between 4d and 5d, and 5d and 6d.

From my experience, I started playing at age of 14, and after playing for 4 months some 5-6 quick games a day with my brother, we became some 5k. (that is 1k / week). Only with playing.
After that, since I did not have proper teacher, I had to work more on theory and read books, and it took me some 1,5 years to reach 1 dan (total of 2 years).
To reach following categories, it took me a year for each dan, and I reached 4d at some 19 (5 years of studying). It could be faster if I had coach, more books, etc.

Next, there was great difference for me between 4d and 5d. That was difference to reach top players in country.
After studying with them, it was necessary to read even more books. I had to move to stronger, Japanese literature, and since I don't know to read it, it was difficult to understand. Finally I realized simple recipe - player who plays last in diagram is not worse in situation, it is draw or he is better.

At peak, being strong 5d (2523) which some might say is pretty close to 6d, I realized just how far it was:
by looking at players with greater rating, I spotted that all of them had either spent some time studying go on Far east, or were from countries with several 6d+. (same as today, btw)
http://www.europeangodatabase.eu/EGD/Find_Player.php (sort by last column)
That meant I would have to spent lot of time studying, playing on strong tournaments, travelling to Japan - basically devoting my life to that for more than a year, which I could not do since I started to work, and had no support from Go Federation. And for that level, you have to be strong at all segments of game, you can not just count on your talent at one segment, it is necessary to study everything. And all that work for some mere 50 rating points.

That is why it is no surprise that so far there was only around 100 players from Europe who reached 6+d level. Of them, majority spent months in Far east studying go, or were trained in strong local centers.

Therefore, when we have a case that someone jumped from 5k to 1k quickly, sure, it is easily possible.
From 1d to 3d? Yes, even without too much reading.
And that someone who was 4d plays like 6+d?
Sorry, sorry, sorry. It is just insult to European best players who kicked their ass of studying to become that strong.
Please bear that in mind when we discuss this case. Lot of work has to be put into it, and it can be easily shown.


This post by Bojanic was liked by 3 people: Gobang, Hidoshito, Simba
Top
 Profile  
 
Offline
 Post subject: Re: “Decision: case of using computer assistance in League A
Post #569 Posted: Sun Jun 17, 2018 2:28 pm 
Lives with ko

Posts: 141
Liked others: 26
Was liked: 89
Rank: 5 dan
Bill Spight wrote:
Of the 50 plays examined in the Metta-Ben David game, I judged that only 12 were difficult enough to be relevant to the question of cheating. Using similar but different criteria Bojanic found 6 important plays by Metta between move 50 and 105, a smaller range of 28 plays by him. He and I agreed on 4 of those 6 plays. My guess is that with even several judges we would have consensus on at least 3 plays.

Bill,
in Metta-Ben David game, I estimated that middlegame is from moves 45-105, which is 60 and not 50 moves. As you are used to hear on forums - why don't you read what was posted?

In middle game section, we agreed on all 4 important moves, but please not that I also analyzed sequences between them.
Regarding additional plays you mentioned, I agree that they were important, but they are IMO endgame+fighting important moves, and therefore have less strategic value then middle game moves. I have no objections on finding important moves in endgame too, you can do it easily if you are interested. BTW please note that 2 moves you added were also Leela's top choice.
And we have also discussed move 137 from the same game. It is even more important than moves you added, since heavy fighting erupts after that.

So, if you are willing, you can send me list of important endgame moves from 3 games I analyzed (since game with Kim was finished earlier), I will make my list, and we can work on them on same way as middle game moves.

Top
 Profile  
 
Offline
 Post subject: Re: “Decision: case of using computer assistance in League A
Post #570 Posted: Sun Jun 17, 2018 3:44 pm 
Dies in gote

Posts: 65
Liked others: 31
Was liked: 55
Jan.van.Rongen wrote:
So I have a question for Ales Cieply who wrote the original "98%" analysis. That was debunked in much detail by Metta's PhD advisor Francesco Morandin, and I have never seen a detailed reply from Cieply to that report from 23 April 2018. So I have a simple qusetion: where is your reply?


Jan, it would be good if you checked your records. I have not written the original "98%" analysis. I even did not have Leela installed at my PC at that time to do any analysis. I did start my own analysis (a completely different one) at the time the case reached PGETC appeals committee.

Top
 Profile  
 
Offline
 Post subject: Re: “Decision: case of using computer assistance in League A
Post #571 Posted: Sun Jun 17, 2018 3:54 pm 
Lives with ko

Posts: 170
Liked others: 14
Was liked: 23
Rank: 6d KGS
Javaness2 wrote:
This post from reddit strikes me as a crock of shite.
Yes, CM could have cheated, but then why tell people about it. Is the allegation that the entire Italian squad was complicit, if so then doesn't that strike you as slightly beyond belief.


People who have done wrong often confide in someone, to ease the burden on their own conscience. This is extremely common. No, I don't think the entire Italian squad was complicit. I think CM is the only one who has done wrong here. His teammate came forward and spoke up, albeit after a short delay, because it was playing on his conscience. That's fine - I forgive them for the delay. They didn't keep it quiet forever.

Javaness2 wrote:
Yes okay, Leela 11 and Leela Zero differences exist, but so what? Why would CM want to cheat with the obselete Leela 11 in the first place? Why not pick 1 of many strong LZ networks?


He did pick one of the strong LZ networks - and not the most up-to-date one, because that would be too obvious. If he was accused of cheating, he could (and was planning on) pointing at move 156 as proof he wasn't. If he was accused of cheating with Leela Zero, then of course this wouldn't have worked. But having such a quick first line of defence on hand for a general "we think you are still cheating" accusation would've been powerful.

Bill Spight wrote:
Also, in your game vs. Metta, if Black 157 descends, White can win the resulting semeai. The play is pretty much a one lane road, so is within the capabilities of a European 4 dan.


This isn't relevant. If he'd been playing for himself, I'm sure he'd have seen it - I certainly saw it, that's why I didn't descend. No one is contesting this. We all can see that it works. What is far more interesting and not at all obvious is what Leela 0.11 and Leela Zero say about it. There is no easy way of finding that this move is such an unlikely and critical blind spot (see criteria 1-4 below). You can check it when told, and check that the refutation when entered manually does make Leela 0.11 see sense, sure, but that's so different from finding it for yourself.

Fenring wrote:
What i dont understand is why it would be a very difficult task for a malicious person to produce such a statement?(Np-problem)
All he have to do is to check the Carlo's moves where Leela 0.11 and Leela Zero disagree? check the moves "Leela 11 says it is very bad move but Leela Zero says it is super so he plays it", i dont know how many of them exists,and on which metrics it is based(top 3 choice,Winrate variation) but it doesn't matter,we are clearly not in a situation "impossible to find,easy to check" like PvsNp problem.


I refer you back to PF137's post here: https://www.reddit.com/r/baduk/comments ... d/e0c509f/. If you play this way, Leela 0.11 believes it's going to lose until manually shown the refutation. That is a pretty big thing that could be pointed to as proof that "I didn't cheat, see, look, Leela says it would lose if I did that". Leela 0.11's suggested move wins the game. This is in the endgame too.

I challenge you to find another situation in any of Metta's online games, PGETC or otherwise, played between the release of 0.11, and the date the cheating accusation was first made where the following criteria are met:

1) We are in the endgame. Let's say move > 150 as a guideline.
2) Leela 0.11's best move shows a significant victory for side A.
3) Leela 0.11 has at least one move that incorrectly shows a significant defeat for side A, and Leela 0.11 does not see that it is incorrect until manually shown the refutation.
4) Leela Zero correctly finds the refutation, and recommends the incorrect move itself as its first choice.

Surely you can see that those criteria, which are all met here, are needle-in-haystack. No one is going to sit there to find something like that. You'd also have to be very strong to find something like this because you'd need to meet criteria 3, i.e. you need to find the refutation, or explore extensively with Leela Zero trying to find blind spots in each position in Leela 0.11. The time required here is immense. In comparison, actually checking the presence of such a move that meets the criteria, if you're told where to look, is very, very easy.

Jan.van.Rongen wrote:
Simba wrote:
As it happens, the move in question (156) was Leela 0.11's third choice, ...


What does that show? Nothing IMO.

For the 5 alternatives White has for this move (h19, J-19-J16) 3 win the game if my analysis is correct. J18 played by Metta is a well known shape move. It is not even too difficult to see that it works when you notice the dame zumari of Blacks bigger dragon.


Lol, what are you talking about? The person that I quoted was harpering on about the whole top-3 Leela thing from ages ago, and how he would, as part of his finding of such a move, exclude top-3 moves by Leela. When in fact the move demonstrated was in the top 3 Leela 0.11 chose. Don't try to use what I've said out of context.

Jan.van.Rongen wrote:
You are a 3 dan who lost to a 4 dan. What's wrong with that? Why put up a show how you let your team down etc. ? Your only "evidence" that white cheated is an unnamed person that made some remarks on reddit.


My only evidence? Please read through the rest of the thread and the other analyses that people have provided.

I apologise in advance if the next section comes off as arrogant; I'm simply trying to state things factually here.

I'm not 3-dan. I'm 6-dan on KGS, and routinely chew up 4d players without any issue (in fact I do paid teaching for players up to 3d level). I've won every game I've played in PGETC (the tournament that these games are part of) for the past two years, and beat the British champion in fairly serious games with mid-long time limits last year 5-0, albeit with one win on time (he's 4-5d). I attempted to reset to 5-dan before the PGETC league started, but the league organiser didn't allow it, so I'm stuck with an outdated EGF rating and beat the stuffing out of most people I get put against on the lowest board (because the boards are forcibly ordered by rating, as per PGETC rules, and my country has several people with a higher EGF rating than me). I've beaten professionals on even before several times and don't feel out of my depth against them.

EGF ratings are completely unsuitable for players who are primarily online-based. It's impossible to get an accurate reflection of someone's skill from half a dozen games per year in a rating system. But that's one hell of a large topic, and one that we shouldn't get distracted by here. My last face-to-face tournament was in 2016, and I lost two games from lol-nonsense in byo-yomi from clearly winning positions because I'm not used to playing in that setting. Perhaps in over-the-board play, while I remain rusty at it, and that's unlikely to change in the near future, I'd only be 3-4d, but online? I give 2-3 stones to those players.

The gap between 4d and 6d is enormous. I don't want this to sound unkind but you really have no perception of how much stronger than 6d - stronger than the professionals that I've played - someone would have to be to make me feel so utterly helpless like in that game. I wasn't playing against a 4d. No chance in hell. It scares me even thinking how many stones I'd need to take from an opponent that strong. Maybe I'd have a chance with four, but I'm sat here now doubting myself even with four stones. But yeah, don't think for a single second that it's somehow normal for me to get blown into space without any glimmer of hope by a 4d. It isn't. And that certainly isn't unique to me; all 6ds will tell you that 4ds just can't completely oppress them to the point of feeling like they have no hope. You can say all you want as a kibitzer, but you weren't playing. You didn't feel that; don't even pretend to know what it was like.

Bojanic wrote:
There is one important point that we have not discussed in this topic.
That is how big difference is between 4d and 5d, and 5d and 6d.


^ This, this so much. And how big the difference between 6d and someone who completely tears them to pieces is. It's pure naivety (and slightly insulting, though understandable given lack of perception of this kind of difference from a player who isn't at this level) to suggest that a 4d is capable of that with no counterplay available.


This post by Simba was liked by: Gobang
Top
 Profile  
 
Offline
 Post subject: Re: “Decision: case of using computer assistance in League A
Post #572 Posted: Sun Jun 17, 2018 4:16 pm 
Honinbo

Posts: 9135
Liked others: 2794
Was liked: 3093
Bojanic wrote:
Bill Spight wrote:
Of the 50 plays examined in the Metta-Ben David game, I judged that only 12 were difficult enough to be relevant to the question of cheating. Using similar but different criteria Bojanic found 6 important plays by Metta between move 50 and 105, a smaller range of 28 plays by him. He and I agreed on 4 of those 6 plays. My guess is that with even several judges we would have consensus on at least 3 plays.

Bill,
in Metta-Ben David game, I estimated that middlegame is from moves 45-105, which is 60 and not 50 moves. As you are used to hear on forums - why don't you read what was posted?


Bojanic, the first important play that you identified was move 51, which came after move 50, the start of the sequence of 50 moves by Metta used in the original verdict, which was where I started looking. You stopped at move 105, which is before the end of that sequence. Sure, there are 60 moves in your sequence, 30 moves by Metta, 28 of which fall in the sequence both of us looked at.

Before you start accusing me of not reading what you wrote, why don't you read what I write more carefully?

Edit: As I said in my response to Ed, I cut you some slack the first time. But not the second.

_________________
The Adkins Principle:

At some point, doesn't thinking have to go on?

— Winona Adkins

"Once in a very great while his eyes light up for a moment, and he says "Whee!" very quietly."
— Lion Miller

Top
 Profile  
 
Offline
 Post subject: Re: “Decision: case of using computer assistance in League A
Post #573 Posted: Sun Jun 17, 2018 4:28 pm 
Dies in gote

Posts: 31
Liked others: 3
Was liked: 7
Rank: 5 kyu
Simba wrote:

The gap between 4d and 6d is enormous. I don't want this to sound unkind but you really have no perception of how much stronger than 6d - stronger than the professionals that I've played - someone would have to be to make me feel so utterly helpless like in that game. I wasn't playing against a 4d. No chance in hell. It scares me even thinking how many stones I'd need to take from an opponent that strong. Maybe I'd have a chance with four, but I'm sat here now doubting myself even with four stones. But yeah, don't think for a single second that it's somehow normal for me to get blown into space without any glimmer of hope by a 4d. It isn't. And that certainly isn't unique to me; all 6ds will tell you that 4ds just can't completely oppress them to the point of feeling like they have no hope. You can say all you want as a kibitzer, but you weren't playing. You didn't feel that; don't even pretend to know what it was like.



I quickly reviewed your game with Leela Zero to check out some of your sentences. Surprisingly I saw that White made a huge blunder with the move 74, and after sacrificing his group with the 76-82 sequence you were in a solid winning position according to LZ.

This is very contradictory compared to your version of being "completely oppress". :scratch:

Top
 Profile  
 
Offline
 Post subject: Re: “Decision: case of using computer assistance in League A
Post #574 Posted: Sun Jun 17, 2018 4:42 pm 
Dies in gote

Posts: 31
Liked others: 3
Was liked: 7
Rank: 5 kyu
Simba wrote:

EGF ratings are completely unsuitable for players who are primarily online-based. It's impossible to get an accurate reflection of someone's skill from half a dozen games per year in a rating system. But that's one hell of a large topic, and one that we shouldn't get distracted by here. My last face-to-face tournament was in 2016, and I lost two games from lol-nonsense in byo-yomi from clearly winning positions because I'm not used to playing in that setting. Perhaps in over-the-board play, while I remain rusty at it, and that's unlikely to change in the near future, I'd only be 3-4d, but online? I give 2-3 stones to those players.



So you accidentally admitted that there may be discrepancy between live and online rating and strenght, so much so that on KGS you can give 2-3 stones to someone who can beat you in live matches.

It is also interesting how you judge your mistakes compared to Carlo's mistakes. You called yours as "lol-nonsense in byo-yomi from clearly winning positions", Carlo's live mistakes are instead "evidence of cheating".

If we were in a court of justice your allegations would be annihilated in zero time.


This post by bugsti was liked by 2 people: Charlie, zermelo
Top
 Profile  
 
Offline
 Post subject: Re: “Decision: case of using computer assistance in League A
Post #575 Posted: Sun Jun 17, 2018 4:47 pm 
Honinbo

Posts: 9135
Liked others: 2794
Was liked: 3093
Simba wrote:
Bill Spight wrote:
Also, in your game vs. Metta, if Black 157 descends, White can win the resulting semeai. The play is pretty much a one lane road, so is within the capabilities of a European 4 dan.


This isn't relevant. If he'd been playing for himself, I'm sure he'd have seen it - I certainly saw it, that's why I didn't descend. No one is contesting this. We all can see that it works. What is far more interesting and not at all obvious is what Leela 0.11 and Leela Zero say about it. There is no easy way of finding that this move is such an unlikely and critical blind spot (see criteria 1-4 below). You can check it when told, and check that the refutation when entered manually does make Leela 0.11 see sense, sure, but that's so different from finding it for yourself.


What I said next was this:
Bill Spight wrote:
OTOH, we know that in the heat of battle mistakes are made.
I was pointing out that it was possible that Metta might have missed that fact, left to his own devices. (Not that I am accusing him of cheating.)

I was talking about analyzing the play in terms of go, not statistics. Which is why I followed by saying this:
Bill Spight wrote:
Such go analysis is like expert testimony in court, not infallible, but not worthless, either.


As for the question of Leela 11 vs. Leela Zero, that was irrelevant to my point. Sorry for not making that clear. But I also indicated in another note that I thought the question of comparing Leela 11 and Leela Zero was a diversion created by the anonymous accuser. I doubt if Metta (or anybody in their right mind) would cheat by checking the choices of both Leela 11 and Leela Zero and picking one of Leela 11's top three choices unless Leela Zero indicated that Leela 11's top choice was a mistake. But that's my opinion. I am not going to argue the point. I am not here to argue. (Even if I seem to be doing a lot of it lately. ;))

_________________
The Adkins Principle:

At some point, doesn't thinking have to go on?

— Winona Adkins

"Once in a very great while his eyes light up for a moment, and he says "Whee!" very quietly."
— Lion Miller


This post by Bill Spight was liked by: Charlie
Top
 Profile  
 
Offline
 Post subject: Re: “Decision: case of using computer assistance in League A
Post #576 Posted: Sun Jun 17, 2018 5:55 pm 
Lives with ko

Posts: 170
Liked others: 14
Was liked: 23
Rank: 6d KGS
bugsti wrote:
I quickly reviewed your game with Leela Zero to check out some of your sentences. Surprisingly I saw that White made a huge blunder with the move 74, and after sacrificing his group with the 76-82 sequence you were in a solid winning position according to LZ.

This is very contradictory compared to your version of being "completely oppress". :scratch:


Look at the rest of the game. When I perform analysis on this myself (and I will, just I don't have much time at all for this sort of thing), I expect to find that he turned LZ on at a certain point - likely after he messed up in this spot.

The idea that a 4d could legitimately make me look like I have no idea how to play from that point is even more eye-watering than the idea that they could do that from start to finish.

bugsti wrote:
So you accidentally admitted that there may be discrepancy between live and online rating and strenght, so much so that on KGS you can give 2-3 stones to someone who can beat you in live matches.

It is also interesting how you judge your mistakes compared to Carlo's mistakes. You called yours as "lol-nonsense in byo-yomi from clearly winning positions", Carlo's live mistakes are instead "evidence of cheating".

If we were in a court of justice your allegations would be annihilated in zero time.


'Accidentally' admitted? One has to wonder what you think I was trying to say if not that there is a discrepancy between the two, in particular for people who play almost exclusively one but not the other. Regardless, CM plays plenty in real life; this isn't relevant in his case.

CM's mistakes are not evidence of cheating, the fact his play matches a bot when he has access to one, and mismatches a bot when he has does not have access to one is evidence of cheating.

In my case, the only difference in my play occurs after my main time is over. I suck at real life byo-yomi since I'm just completely not used to it :P . I panic and still play stones out of my bowl instead of out of the lid of the container while blitzing everything at 2 seconds/move because I'm worried about losing on time. When not in byo-yomi, I'm as strong face-to-face as I am online.

Top
 Profile  
 
Offline
 Post subject: Re: “Decision: case of using computer assistance in League A
Post #577 Posted: Sun Jun 17, 2018 10:37 pm 
Tengen

Posts: 4824
Liked others: 0
Was liked: 656
Simba wrote:
The idea that a 4d could legitimately make me look like I have no idea how to play from that point


You rely on the wrong assumptions that your level of play is constant, the level of play of a 4d opponent of yours is constant, a 4d cannot play like a 6d in some of his games or some parts of his games, a 4d cannot create positions whose treatment are your weakness.

Quote:
The gap between 4d and 6d is enormous.


The gap in effort to reach the ranks - yes. The gap in explicit or subconscious knowledge - yes. The impact on play in a single game - no (as a 5d, I know). The major impact on play in a single game is the difference in winning probability when continuing from the same position: my 4d opponents lose more often and my 6d opponents win more often from the same kind of position. The gap in explicit or subconscious knowledge can be noticed on average over several games against the same or different opponents but does not enable the 6d opponents to play better than 4d in every game and on every move. Proof: 6d (and even 7d) opponents do lose a significant fraction of their games against 5d players.

Furthermore, I have seen great differences of particular players' skill in online versus real world games. Some are better online, some better in the real world.


This post by RobertJasiek was liked by: Charlie
Top
 Profile  
 
Offline
 Post subject: Re: “Decision: case of using computer assistance in League A
Post #578 Posted: Sun Jun 17, 2018 11:09 pm 
Lives with ko

Posts: 141
Liked others: 26
Was liked: 89
Rank: 5 dan
Bill Spight wrote:
Bojanic, the first important play that you identified was move 51, which came after move 50, the start of the sequence of 50 moves by Metta used in the original verdict, which was where I started looking. You stopped at move 105, which is before the end of that sequence. Sure, there are 60 moves in your sequence, 30 moves by Metta, 28 of which fall in the sequence both of us looked at.

Before you start accusing me of not reading what you wrote, why don't you read what I write more carefully?

Bill,
in updated paper, given here:
viewtopic.php?p=232713#p232713

and in same chart posted earlier here:
viewtopic.php?p=232557#p232557

You can see in yellow analyzed sequences.
Please note that I marked that "joseki" ended at 38, and that sequence of moves were analyzed after that.
There is no Metta's tenukis in this sequence, since he replied to his opponent's moves.
You can also note that chart is first version - I marked move 45 a tenuki too, but during detailed analysis I realized it was just a reply.
Therefore, moves from 38 to 105 were analyzed, not 50 to 105, which is clearly (marked yellow) visible in chart shown two times in this topic. Same situation is visible in other games.

Top
 Profile  
 
Offline
 Post subject: Re: “Decision: case of using computer assistance in League A
Post #579 Posted: Sun Jun 17, 2018 11:22 pm 
Honinbo

Posts: 9135
Liked others: 2794
Was liked: 3093
Bojanic wrote:
Therefore, moves from 38 to 105 were analyzed, not 50 to 105, which is clearly (marked yellow) visible in chart shown two times in this topic.


I never claimed that you analyzed only moves 50 to 105, I said that that range contained the moves that you analyzed and the moves that I analyzed. We analyzed different ranges.

_________________
The Adkins Principle:

At some point, doesn't thinking have to go on?

— Winona Adkins

"Once in a very great while his eyes light up for a moment, and he says "Whee!" very quietly."
— Lion Miller

Top
 Profile  
 
Offline
 Post subject: Re: “Decision: case of using computer assistance in League A
Post #580 Posted: Sun Jun 17, 2018 11:40 pm 
Dies in gote

Posts: 63
Liked others: 0
Was liked: 12
Rank: FFG 5k
Simba wrote:
Fenring wrote:
What i dont understand is why it would be a very difficult task for a malicious person to produce such a statement?(Np-problem)
All he have to do is to check the Carlo's moves where Leela 0.11 and Leela Zero disagree? check the moves "Leela 11 says it is very bad move but Leela Zero says it is super so he plays it", i dont know how many of them exists,and on which metrics it is based(top 3 choice,Winrate variation) but it doesn't matter,we are clearly not in a situation "impossible to find,easy to check" like PvsNp problem.


I refer you back to PF137's post here: https://www.reddit.com/r/baduk/comments ... d/e0c509f/. If you play this way, Leela 0.11 believes it's going to lose until manually shown the refutation. That is a pretty big thing that could be pointed to as proof that "I didn't cheat, see, look, Leela says it would lose if I did that". Leela 0.11's suggested move wins the game. This is in the endgame too.

I challenge you to find another situation in any of Metta's online games, PGETC or otherwise, played between the release of 0.11, and the date the cheating accusation was first made where the following criteria are met:

1) We are in the endgame. Let's say move > 150 as a guideline.
2) Leela 0.11's best move shows a significant victory for side A.
3) Leela 0.11 has at least one move that incorrectly shows a significant defeat for side A, and Leela 0.11 does not see that it is incorrect until manually shown the refutation.
4) Leela Zero correctly finds the refutation, and recommends the incorrect move itself as its first choice.

Surely you can see that those criteria, which are all met here, are needle-in-haystack. No one is going to sit there to find something like that. You'd also have to be very strong to find something like this because you'd need to meet criteria 3, i.e. you need to find the refutation, or explore extensively with Leela Zero trying to find blind spots in each position in Leela 0.11. The time required here is immense. In comparison, actually checking the presence of such a move that meets the criteria, if you're told where to look, is very, very easy.

Simba, as i try to explain you since 2 posts, this is obviously not needle-in-haystack, and logic of PF137 you already show is really flawless.


Someone who guess CM cheat with Leela Zero can easily find this.
I open Carlo's Game with Leela Zero and Leela 0.11.
I take a look only when the moves played by Carlo is first choice of Leela Zero and have lower winrate with Leela 0.11.
On this very few moves, i can investigate further,and i dont even need to be strong to check the 3). Leela Zero will give me the refutation.

And i repeat the process on all games i want to check(those played by Carlo after Leela Zero was strong enough).
Not really a needle-in-haystack. i just open the game with 2 bots instead one and compare.

Top
 Profile  
 
Display posts from previous:  Sort by  
Post new topic Reply to topic  [ 720 posts ]  Go to page Previous  1 ... 26, 27, 28, 29, 30, 31, 32 ... 36  Next

All times are UTC - 8 hours [ DST ]


Who is online

Users browsing this forum: No registered users and 1 guest


You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
You cannot post attachments in this forum

Search for:
Jump to:  
Powered by phpBB © 2000, 2002, 2005, 2007 phpBB Group