you realize they're not betting real money right? and the humans are freerolling a $100k prize? and the bot plays on its own? cuz it's a bot?

if the bot is down enough shouldn't they call it quits anyway?

A developer was in twitch for a few minutes.

Noambrown: Claudico developer here. I have a few free moments so I'm happy
Noambrown: hi
Knorrli: yo noam, does claudico consider timing tells?
Noambrown: It doesn't consider timing tells; we told the pros this
Noambrown: We didn't want them to take 3 minutes every hand... like us
Nosedoto: NoamBrown, can you speed up the river decisions with like memcached or something?
Noambrown: We have a special process that computes a strategy for the river on the fly. It takes a while to solve, up to 3 mins
Alall4: is the bot trying to play gto type strategy like cepheus or it is it developing its strategy around something else
Noambrown: We don't want to give too much of the bot away, but we do use regret minimization algorithms, as well as some other things thrown in the mix
Noambrown: Claudico decided on the 10% river pot size. I don't have a say in
Noambrown: There's 2 students on the team
Noambrown: There's no neural net going on, but you should expect the bot to change
Noambrown: It will be playing differently tomorrow
Noambrown: We compute the strategy on about 2K cores and about 5 TB of memory. The river is computed on 64 cores and 512 GB memory
Noambrown: The strategy was computed over months.
Noambrown: I'm pretty happy with how it's doing so far. There's clear weaknesses but I think it will improve
Noambrown: Bots can play multi table but they aren't very good at it
Encryption: Some sort of grant/funing for this project?
Noambrown: They're using hotkeys to change the numbers.
Noambrown: The pros get the money either way, and we raised it through sponsors

lol

They've been ribbing each other back and forth at any given chance. This was just playful banter and not a serious concern or complaint.

It's kind of funny that they are exploiting the bot by donking close to 100% to make the game tree unfamiliar for the bot, as I doubt that's a legit strat (?). : D

lol @ the table design...looks like they put about an hour into it. Really hard for a viewer to follow the action, the chips/betting placements make it look like a casino game.

so the current winrate is about 26bb/100 if im not mistaken. a few more sessions with this winrate and they can call it gg, unless they get extra money the more theyre up. if the winrate does not decrease it seems pointless to play 80k hands imo

Polk seems to have applied a style best suited to exploit what he knows of the bot's weaknesses whereas Jason Les seems to .... well, be a bit of slow learner.

Perhaps Polk will coach him up overnight for Team Human.

I've never heard of the other 3 guys. They also play high stakes?
I think it would of been more fun to see a average player play against this machine.

Yes, even a 1 second pause before the bot acts (other than the river where it takes a minute or so) would make the game easier to follow.

What would that prove? Define "average."

Jason Les is playing poorly enough that I think he qualifies. Enjoy.

Pretty sure that for the programmer's team it's not "pointless". They are allowed to change the bot, and will do so, to improve the AI and learn for the future. I would be disappointed if after forur days or w/e they threw in the towel, because decreasing the human's WR significantly over the full course of the match will be a great achievement by itself.

Well Doug was running quite hot the first session and I think the bot Jason is playing is getting WCG's cards. So that might be part of it. Still early in the match.

so cute floorman to ask for doug to help him out with some hands

edit: lol next floorman stands with cardplayer ready, probably to get wcg to sign it if he have been in it (very likely)

Basically from what I've seen so far the exploits are:

-fold vs bot's huge overbets
-make cheap bluffs oop
-overcbet flops and don't worry about unbalanced check back range ott

At least that's what wcg has been doing.

I think the humans are already getting both sides of the cards each round: Upstairs humans are getting bot's downstairs cards and downstairs humans are getting bot's upstairs cards. Jason Les, from the hands I saw, is just playing very poorly and not making the same insights that Doug did.

But if the programmer can adjust the Bot and so on for the next day, then u cant call it AI can u? I mean the bot doesnt have to adjust to the game by himself and not by the programmer. Or do i miss something?

1)River computed on 64 cores with 512GB memory - still tanks every time the river is reached.
2)Strategy computed over months on 2k cores - unexpected minbetting OOP forces necessary adjustments overnight

There was also a moment when Doug was trying to explain the basics of nut/bluffs vs bluff catchers game to the author, it didn't go so well ("I don't have many jacks in my range so it payoffs to bet bigger with value and bluffs vs me", "but you wouldn't call a bigger bet, would you?").

I mean srsly.. all this doesn't add up. It looks like huge misunderstanding to me.

Les has been getting the same cards as bot´s getting against WCG and is down 70k, bot is down 115k so not so bad by Les. Still a problem though beein down so much against bot.

^are those numbers over the span of 750 hands?

I'm not very surprised. I beat the neopokerbot for a similar winrate over ~13k hands and was experimenting quite a lot before I settled for an easy profitable strategy. I think that if you gave Team Human 3k practice hands, an interface like stars or tilt, instant reply instead of these 3 min river tanks, better service (like not having them beg for water) and a secluded area (not in the middle of a casino where people shout and supervisors and programmers blabbering in their ears you would see winrates closer to 40bb/100, maybe more.

The bot is no different than other bots. It has a hard time vs bets that aren't in the vicinity of 1/3pot-1pot, it doesn't seem to utilize card removal, it bluffs badly and huge bets are nuts. I mean it jams for 19x pot and shows up with second backdoored nut straight, how can that be a part of any sound strategy?

They each played 2 stretches of 750 hands, so 1,500 hands.

Yeah, I was also thinking this - invest some more programming hours to create a decent interface, and let them 2-table. 2-tabling a "nice" interface would, even for poker novices, probably be easier to follow than 1-tabling the actual interface, and the volume would come in much quicker. Match over after one week instead of two -> win-win.

Also les isn't even close to being one of the best HU players in the world, when they (evil empire) was doing those power rankings at no point did his name come up in top 20.

Why isn't Klink playing instead of les?