You guys are harsh.

Yes claudico has some faults, and yes it's surprising they aren't calculating exploitability.

The faults it has should be patched relatively easy. I doubt we'll need to see it play next year to know it will crush humans.

Thank you for that and I apologize for leaving out the word "fairly" as I just heard it live and was stunned by the "fourth" comment after it that my recollection was affected.

This event shouldn't be about beating humans so there is no shame that it cannot do that yet (to me it is about improving AI by using the test of poker), I just thought it was rather small to re-frame the Event like he did while ignoring the mirroring of hands that make ranking irrelevant and making Jason look bad when in fact his position relative to the others may largely/exclusively be due to the cards. The Prof can sell the computer any way he wants to the press, but Jason's rep paid the price for that exaggeration/mischaracterization which is indefensible.

Imagine if the Prof had said, "We have the best AI computer program in the world and yet these Pros are overall crushing it which really is a testament to their abilities and how much we still have to learn. But, the event is still not over and the computer is learning more every day so who knows how it will end."

Would be really interesting if they had 15k left for a second challenge with decent HU pros from 100-400nl.

I'd definitely bet on the bot then, especially if they lower the stacks to 100bb. And even more if the HU players could lose money. Make it 1/2 or 2/4 blinds with real money and claudico wins.

Honestly, we had just assumed computing exploitability is hard, since we use imperfect-recall abstractions and there aren't any known approaches for computing best responses in imperfect-recall games. Your earlier post suggested maybe we could still bound exploitability by computing a best response to our strategy in a different abstraction that had perfect recall (and probably just one bet size). It's not clear to me if it's doable, and what the scientific value is of computing a bound on our exploitability like this. When I have some actual time (i.e., after I have my diploma in hand) I'll think about it some more and get back to you.

I'm heading in today. Post some questions that you want me to ask the players on twitch and I'll ask as many as I can.

Did I say there were tweaks?

Would love to know how far wcg thinks he is from a perfect gto strat, in terms of winrate..

Apparently you did not, I apologize, I know you hate people attributing quotes to you that wasn't yours . I could have swore one developer was in this thread stating something to that extent though.

I found this:

So my question still remains I guess, why would the bot play different from one day to another if I'm right in my assumption that its strategy is based on previous hands played is correct since the hands played during the span of one day should be like a drop in the ocean.

Questions for player:

What tactics used in the event were applied by you specifically because you thought they would be more successful from the get-go against a bot than a good human player?

Are you intentionally making less-than-optimal strategy/style choices because you are on stream and know future human competitors are watching how you play?


Thanks.

Ask them what part of their winrate against the bot comes from the exploitative adjustments they have made during the match.

Thank you Sam.

You remove bjorn li(asian bot) from current stats

and the bot's doing pretty amazing.. 17k vs dong and only losing 6stacks is impressive. If they were playing 100nl online with rake the bot dong+doug+jason would be breakeven right now vs the bot

Am I missing something? I have Dong + WCG + Cheet up ~ 207K

Removing anyone when the hands are "mirrored" is less than pointless.

He seems to be trolling and have a hate-on for Jason.

No strat until after Friday bro

Got ya

I've always loved the world of nerds, when you're around them anyone with any kind of opinion is a troll or a bully.

If you factor in rake, those 3 put together are breakeven at any stake under 400nl. Bjorn is the only one doing decent against the bot. 17k hand is a pretty decent sample when you factor in that those guys are the top of the pool.

Yet people are still saying that claudico is a clear failure when a lot of 200 and under reg are making good money beating on other regs while paying a lot of rake. Top highstakers would be losing/breaking even vs this bot in the same situation. The truth is that claudico is already a ton better than almost every reg.

Doug's down over 100k today through 400 hands, and appears to be playing some disinterested poker. Fatigue looks to be a factor.

Do you know what the rake is at "highstakes" relative to the stakes? Why would "top highstakers" drop down to be in the "same situation" as the 200 and under regs you reference to play this bot? If you are trying to say they would not be winning as much if they were also paying rake, well, duh.

As for the world of nerds, your desire to group players to support a point does not make you a troll or bully to them; simply, misguided.

The mirroring of hands IS significant which is why, if you want to group players to draw a conclusion, you would look at the grouping of Bjorn and Dong who are playing mirrored hands. Both players are winning quite a bit last I looked with the computer playing each others' hands. Now if the bot is able to end up on both Doug and Jason then I would consider that a significant achievement and one that will be noticed by those in the world of poker and AI alike.

I don't see how its so hard to understand that if highstakes regs can't beat this bot at lower stakes online, its a pretty amazing bot.

98% of people living from poker are paying high rakes so making comparisons with raked environment is a must to gauge the quality of the bot.
if you drop a top hu pro vs a 100nl hu player @ 100nl. The top pro would win a decent ammount long-term and now they are breaking even vs claudico if you factor in the same rake.

In fairness, the professor is likely suffering from something that many professors are afflicted with early in their career: tenuritis. This manifests itself with all manner of inexplicable symptoms, including chronic arrogance.

Given how things are going there is a slight chance that Claudico will at least reach a statistical tie in the Doug&Jason-part of the competition.

I'm sure Claudico will improve alot in a couple of years and come back stronger. Which makes me happy that I'm a 6-max player.

Doug really seems tired and a bit in idc mode and just wants it to be over. Folding 3k hands should be -225k fwiw . He had periods were he was just rushing through hands on the first days, too, but I feel like his attitude has really changed (starting yesterday).

Nice interviews with Dong and Jason, Sam. Much more enjoyable than muted Doug/Bjorn streams.

@cheet
hand #207 7s6s vs AcQc where claudico calls your chk/jam on Jc8s4s in a 4bet pot you would've rivered the straight flush lol, turn Qo.

The four players in this contest are representing poker very well, all bright and articulate as opposed to the stereotypical poker player. This is good for poker. Poker is a game of skill and needs to differentiate itself from games of pure chance, this is helping to make that distinction to the general public.

The object of this AI experiment, as I understand it,is to program the computer to be able to teach itself how to play poker in NE not to program specific human divined strategies to be the best poker player. This is an important distinction yet I'm getting mixed messages from the AI team about their input in to the program regarding strategy. As I understand it the whole idea is for the computer, by itself, to learn how to solve complex game theory to be used in other areas not to specifically program strat.

So if NE is the aim of the computer it should be apparent there are many spots where it is obviously not in NE. Does the AI team not see this? They seem too focused on results instead of analyzing and seeing where Claudico is relative to NE. I believe Claudico is a very good HU player, but the best HU player may not be anywhere close to NE, nor should they.

At this point, I think there is a much bigger than slight chance that Claudico gets a statistical tie vs Doug and Jason.

Update on the totals please!