WCGRider, Dong Kim, Jason Les and Bjorn Li to play against a new HU bot - Page 9 - Poker News

The programmer seems surprisingly uninformed about poker theory. Doug had a river spot where he was facing a 2x overbet on a 4 straight board and he commented that he liked the bot's bet sizing because he would rarely have a straight but the bot could have a lot more straights. The programmer responded in confusion asking why the bot wouldn't just bet smaller to try to get called if it had the straight, and Doug semi-seriously asked if he was leveling him.

Quote

04-24-2015 , 06:34 PM

#202

+VLFBERH+T

grinder

Join Date: Mar 2013 Posts: 657

Quote:

Originally Posted by MaxRhino

So the stack sizes reset to 20k every hand?

What does everyone think about that? Doesn't really track to how a normal HUNL match would play obv. Does that matter though?

That's the same format that the computer competitions are played.

I guess with not re-setting, human edge would be bigger, because it is harder to train the bot to all kinds of different stack sizes too.

Quote

04-24-2015 , 06:36 PM

#203

Dantes

Carpal \'Tunnel

Join Date: Jun 2007 Posts: 10,833

Quote:

Originally Posted by NVF

I thought the same thing.
"He has a stronger range so he wants the pot to be bigger"
"why?"

Quote

04-24-2015 , 06:41 PM

#204

Chabra

grinder

Join Date: Apr 2013 Posts: 403

Quote:

Originally Posted by MaxRhino

So the stack sizes reset to 20k every hand?

What does everyone think about that? Doesn't really track to how a normal HUNL match would play obv. Does that matter though?

I think it's good, sets a standard. Also it's better to do it this way is because one player is getting the reverse cards of the other player. So if they didn't reset the stack and say there was a huge cooler, one player might win 200bb and the other lose 500bb and that would somewhat defeat the whole purpose of trying to reduce/ eliminate variance.

Quote

04-24-2015 , 06:45 PM

#205

+rep_lol

banned

Join Date: Sep 2013 Posts: 14,900

this is super cool

Quote

04-24-2015 , 06:47 PM

#206

MastaAces

Carpal \'Tunnel

Join Date: May 2007 Posts: 10,007

Quote:

Originally Posted by NVF

Doug had a river spot where he was facing a 2x overbet on a 4 straight board

doug referred to this as 1 to a straight. that's not common nomenclature is it? i thought i knew of all the nerdy terminology.

Quote

04-24-2015 , 06:54 PM

#207

RobFarha

Carpal \'Tunnel

Join Date: Feb 2010 Posts: 9,684

is it now streaming now?

Quote

04-24-2015 , 06:59 PM

#208

doctor877

veteran

Join Date: May 2014 Posts: 3,062

First session (750 hands) already done, another one starting in like 30-60 min afaik.

Quote

04-24-2015 , 06:59 PM

#209

+VLFBERH+T

grinder

Join Date: Mar 2013 Posts: 657

Quote:

Originally Posted by RobFarha

is it now streaming now?

They are having a break, half an hour I think.

Quote

04-24-2015 , 07:02 PM

#210

TooCuriousso1

Golden 2017

Join Date: Jul 2008 Posts: 6,711

Quote:

Originally Posted by theskillzdatklls

from what i was told they get a flat fee regardless of winning or losing.

yeah its an "appearance fee" because the casino wouldnt let them actually play 25/50. but there is some incentive pay out wise if you do better than others.

Quote

04-24-2015 , 07:04 PM

#211

ArtyMcFly

Carpal \'Tunnel

Join Date: Dec 2014 Posts: 13,256

Quote:

Originally Posted by f0ld44

are they seriously playing for 80k hands? This is gonna take weeks at this pace

20,000 hands for each human. The bot plays the same hands four times.

Official site/blog: https://www.cs.cmu.edu/brains-vs-ai

Quote:

Originally Posted by tultfill

bot losing at 33bb/100.

I'm pretty surprised it's losing so heavily. I thought it would be a lot closer. I've not seen much of the streams. Has Doug noticed/mentioned any obvious imbalances/leaks in the bot's play?

Quote

04-24-2015 , 07:06 PM

#212

RobFarha

Carpal \'Tunnel

Join Date: Feb 2010 Posts: 9,684

Quote:

Originally Posted by doctor877

First session (750 hands) already done, another one starting in like 30-60 min afaik.

Quote:

Originally Posted by +VLFBERH+T

They are having a break, half an hour I think.

thanks

Quote

04-24-2015 , 07:19 PM

#213

loosefoot

grinder

Join Date: Nov 2009 Posts: 463

Quote:

Originally Posted by ArtyMcFly

I'm pretty surprised it's losing so heavily. I thought it would be a lot closer. I've not seen much of the streams. Has Doug noticed/mentioned any obvious imbalances/leaks in the bot's play?

Doug and Dong are both min bet leading 100% of flops OOP when defending pre vs a single raise. Doug has mentioned a few times that the computer is not coping well with it.

Quote

04-24-2015 , 07:35 PM

#214

restorativejustice

veteran

Join Date: Aug 2014 Posts: 2,753

The programmer/host chatting with Doug is either extremely uninformed about poker or comes across as such because English is not his first language.

From watching Doug's hands, the bot appears very exploitable by even an average thinking player. You can see the the programmer was crushed when Doug rated that bot a 4 or 5 out of 10 if it were a human player sitting in on his regular tables (His default was not to accept that his bot was flawed but to note Doug's run-good and to imply that Doug must be playing with comparatively great players for his bot to rate so low*).

*In before DN would have trouble beating this bot online.

Quote

04-24-2015 , 07:55 PM

#215

Masq

Carpal \'Tunnel

Join Date: Oct 2008 Posts: 15,008

It's going to be very interesting in about 5 days or so to see how capable the computer is of learning, the first few days I don't expect the computer tp have much chance of winning. Only problem is that the computer may learn how to beat one strategy and then unable to adapt to something completely different the players start to do.

It's back on now!

Quote

04-24-2015 , 07:55 PM

#216

what is check

journeyman

Join Date: Jan 2011 Posts: 373

the funny thing is, i feel like Doug was actually trying to be nice by saying the bot would be a 4 or 5 out of 10.

Quote

04-24-2015 , 08:11 PM

#217

The4thFilm

Carpal \'Tunnel

Join Date: Jul 2004 Posts: 13,472

Is the computer "learning" during this challenge at all?

Quote

04-24-2015 , 08:16 PM

#218

RobFarha

Carpal \'Tunnel

Join Date: Feb 2010 Posts: 9,684

Doug wrecking this thing

Quote

04-24-2015 , 08:18 PM

#219

Masq

Carpal \'Tunnel

Join Date: Oct 2008 Posts: 15,008

Quote:

Originally Posted by The4thFilm

Is the computer "learning" during this challenge at all?

Yes, that's kind of the point of this.

Quote:

HOW DOES CLAUDICO MAKE DECISIONS?
Claudico performs real-time reasoning during a hand. It also improves its strategy during the match by computing continuously on the Pittsburgh Supercomputing Center’s Blacklight computer. The research team built Claudico using algorithms that analyzed poker rules to devise a winning strategy. Claudico is not based on the experience of expert human players, so its strategy for playing can differ markedly from seasoned pros.

Quote

04-24-2015 , 08:36 PM

#220

Zenzor

Pooh-Bah

Join Date: Dec 2007 Posts: 5,420

stream quality is ****ing awful

Quote

04-24-2015 , 08:45 PM

#221

The4thFilm

Carpal \'Tunnel

Join Date: Jul 2004 Posts: 13,472

Quote:

Originally Posted by Masq

Yes, that's kind of the point of this.

I meant learning as in interpreting winning lines from his opponents and copying them. What they're describing is just brute force which they would be doing anyway.