Won a game against a 2p with LeelaGPU on FGS + commentary

[tartaric]

Hi everyone,

Some of you may wonder how strong Leela 0.10.0 GPU could be with a strong GPU? I just had the chance to test it against a pro (Oomote Takuto aka "Demon8528") yesterday on Fox Go server. This pro is not bad, he won against Luo Xihe 9p aka "steadfast" and the legendary Bigticket from KGS (who is 2p) for the older who remember. About the time settings, it was 1 min + 3 periods of 15s/move byo-yomi.

So here is the game with a complete commentary I have done. I hope you appreciate it and let me know what do you think about this victory, the commentary, etc. Personnaly, I didn't think Leela would win and was impressed by his endgame too which wasn't chaotic at all, it was even for some moves in the end very good!

https://gokibitz.com/kifu/SkOOkZIK-

[Cyan]

What are your computer specs? Did you have to relay or does fox support GTP?

[pookpooi]

Did he know that he play against the bot?

[tartaric]

What are your computer specs? Did you have to relay or does fox support GTP?

GPU (the most important I guess): NVIDIA Quadro P5000
Pascal architecture
2560 CUDA cores
288 GB/s
16GB memory

CPU: XEON E5 2623 v4 8Q (I guess there is a better CPU but we can't get everything haha)
RAM: 30 GO

I had to relay, I dunno if it supports GTP but I like to relay, imho you get more into the game and can gain some benefits, as a player, just from looking.

Did he know that he play against the bot?

Unfortunately not because I didn't think I would play a pro at 8D level (the account will be 9D in 6 victories), but I realised after a few moves that there was the "P" after his nickname. And anyway I don't speak Chinese and they, for most of them, don't speak English. Do you think knowing he was playing against a bot it would have changed something? He didn't try to trick me imo, he played normally.

[alphaville]

Cheating like this is disgusting, you should be ashamed of yourself.

[Dontbtme]

Cheating like this is disgusting, you should be ashamed of yourself.

I'm gonna assume this is a joke given your alias
If it isn't, though, and if tartaric should be ashamed of himself, I wonder how Aja Huang should feel after decimating all these top pros around last New Year's Eve thanks to Master!
ROFL

[Uberdude]

how Aha Huang should feel after decimating all these top pros around last New Year's Eve thanks to Master!

Seeing as decimating only means killing 1 in 10, could I suggest massacring, annihilating or slaughtering as more appropriate verbs.

[Dontbtme]

You definitely could and should though you graciously overlooked the most preposterous mistake (which I sneakily corrected since. )

[Tapani]

Interesting that people don't find it unethical to use computer assistance in online go.

I think it is, and how it might ruin online (and offline) go.

To compare the situation with chess, where computer assistance is a huge problem:

At chess.com, one of the biggest online chess sites the estimation is that 5-10% of the players cheat using chess engines (even if their new cheat detection team seems to do a good job reducing this number). The distribution is not even, less cheating at lower levels and more the higher you get.
At some point they even had to give a bump to the ratings of all titled players with 500 points (equivalent to ~5-10 stones), according to rumour because cheaters made it impossible for masters to gain rating (even international masters (~1p) had some trouble getting a 2000 (~1d) standard rating).

And there are punks making youtube videos on how they checkmated the world champion (by letting their computer play when they were lucky enough to get a game with Magnus).


In online go, this situation could mean that there will be hordes of DDK kids having high dan ratings, and professionals having difficulty getting above (say) 5d, because they lose games against said kids. Maybe the go programs are not really that strong yet, but in a few years they will be.
At say SDK levels, the experience might appear as getting very uneven opponents. Sometimes playing against a DDK and the next game against a professional.

Let's just hope it won't get that bad.

[Dontbtme]

Okay, so first of all, let's be clear: tartaric didn't cheat: he happens to be a super nice guy—an altruist, even—who kindly gave some of his time to help poor little Leela to play on FGS because, you know, without him she wouldn't be able to do that and would live a very boring life indeed
Second, they created a 10 dan rank on FGS I believe, because Fine Art—which happens to be a bot too, probably also using a human collaborator/traitor/cheater to be able to play there —was already at least a stone stronger than pros... but so what? most pros are still 9 dan on FGS, and now that there is a new rank available, some top human pros (like Ke Jie for instance) could prove themselves to be of a higher caliber than the rest of their peers.
And if tomorrow new bots appear that are even stronger, well, instead of deciding that only the Top God of Bots should be 10 dan and that we should drop everyone else's rank as a consequence, we could just add an 11 dan rank, and a 12 dan one, etc. Sure, humans probably wouldn't be able to reach said ranks, but who cares? It's not like they were able to either without bots around—apart from some genius weirdos, maybe, like Dosaku and who knows who .
Anyway, sure: if someone creates an account to play with his bot, then the account's strength will be the bot's one, not the human's one. I agree with that, but... so what?
We're not talking about a tournament where someone could win money or whatever... we're not talking about someone playing at an undervalued rank on purpose either, since after creating a new account strong players (humans or bots alike) have no other choice but to climb their way up from a certain rank (3dan?) before reaching their legitimate one (unless they're pros who want to declare themselves as such, in which case they will be granted the (p) or 9dan rank directly, though they won't always do that—far from it).
But in the end, chill out! Once the bot stabilized at its true rank, then ALL IS WELL WITH THE WORLD

[Bill Spight]

Interesting that people don't find it unethical to use computer assistance in online go.

IMO, it is not unethical if the bot has its own handle, so that there is no pretense. Maybe in the future go servers will have two or three types of handles, one for humans, one for bots, and one for a combination.

In online go, this situation could mean that there will be hordes of DDK kids having high dan ratings, and professionals having difficulty getting above (say) 5d, because they lose games against said kids. Maybe the go programs are not really that strong yet, but in a few years they will be.
At say SDK levels, the experience might appear as getting very uneven opponents. Sometimes playing against a DDK and the next game against a professional.

I think that the handicap system in go will take care of DDK kids with high dan ratings, as they will be, say, 13 dans while human pros are 9 and 10 dans. True, people could still cheat by letting the bot choose some plays. But however unethical that may be, if the cheating is consistent then other players can still get accurate ratings.

[pookpooi]

I agree that [in ranked match] the problem lies into there are more than one being per account. If that account has been used by just only Leela all along then it's fine. But if that account composed of Leela and Human or Leela and Zen, or even me and Ke Jie in the same account, switching to play from game to game then it's not. This inconsistency won't affect just the opponent, but to all players in the server.

[Knotwilg]

I don't see a problem. If I want to know how strong I am and I play online, what do I care if my opponent is a bot or a human?
And if you want to know how strong your program is, what do you care if your opponent is a bot or a human?

The only problem I see is when people start teaching and ask money based on their boosted rank.

[moha]

IIRC a decade ago there already was a clear expectation on chess servers that computer accounts should be marked as (C) or similar. IMO this is all that matters (for combined accounts as well).

The ranking systems should be able to adjust, the problem is not computer accounts but the situation where new accounts average strength could be off. But again, this can be taken care of.

[mhlepore]

IIRC a decade ago there already was a clear expectation on chess servers that computer accounts should be marked as (C) or similar. IMO this is all that matters (for combined accounts as well).

The ranking systems should be able to adjust, the problem is not computer accounts but the situation where new accounts average strength could be off. But again, this can be taken care of.

I was thinking same thing - there must be something on this in the (FGS) Terms of Service.