why should he try playing poker when he can confidently walk around with a fresh perspective in which many rly good professional poker players who have studied poker extensively really have no clue about some of the things he is discussing.

He is in an awesome position and how he decides to use his knowledge is completely up to him, however if he is capable of having a bot that can beat hunl players. this should be a pretty big signal to anyone and a wake up call to many people still playing online poker.

Actually, users of HU bots based off game theory should be one of the easiest to detect from game-play history alone (in the same way as people using bots to play Chess and Backgammon are detected). Very good exploitative bots will be much harder to detect though.

Juk

If only everyone else followed suit ... then I could troll to my heart's content and nobody would care..

I could say "None of man's inventions labeled artificial... flowers.. turf.. sweetener.. leather.. or the thousands of others...even begins to approach the quality of the real thing, but AI will be different."

Hello everybody,

I represent Neo Poker Lab and I’m going to answer to some of your questions. Below you will find an answer to the question we’ve received from one of 2+2 member, specifically “We want to know if you have solve the HU limit holdem or not”

The game is solved numerically. We get to the level of precision high enough to consider FLHU solved. There is no problem for us to increase level of precision say by one order. However it’s not evident if this is practically reasonable because it takes some time plus there are some more interesting problems we have that can be solved by algorythmes used to solve FLHU, including ones not related to poker.

No approximation or Monte Carlo simulation, as well as CFR was employed during calculation. The algorithm searched every possible board cards combination, pocket cards and found optimal action for every possible game history. For this strategy we didn’t use a supercomputer or anything similar.

Also I would like to thank FullyCompletely for his activity and help with explaining current state of research in poker AI field. Also we really appreciate Alberta CPRG contribution to Game Theory and team poker related publications.

Unfortunately Neo is a commercial project so we couldn’t go into detailed discussions about our approach and prefer to keep this confidential as much as we can to protect our know-hows.

Thanks for posting here NLP support .
Why this great result doesn't appear on your website ?
( it will prove that you belong to the neobotpoker team)

I've invested a few years ago (with some partners) in a similar project done within the framework of Central/Eastern Europe universities collaboration, and our lab reports considerable progress. It's a purely commercial project, I must add.

So are you claiming that, for example, your best bot couldn't be beaten for more than 2 milli-big-blinds per hand? If it could, no-one would believe your claim to have solved HULHE.

What about your ACPC entry this year? Do you believe that it too cannot be beaten for 2 mbb/h?

Based on the email address used to register the account, this appears to be legit.

@NPLSupport : Can your AI beat the best humans at NL Holdem HU ?

In your first statement you say you solved the game due to a "level of precision" that you "consider" to be "high enough", which implies that there is a small possibility you considered incorrectly, and the level of precision is not high enough, and therefore the game is not actually solved.

In the second statement you say "No approximation" was employed during calculation.

These statements seem to contradict each other.

I might be misunderstanding what you mean by "solved". Is there a non-zero possibility that another strategy could defeat your strategy? Yes or no?

the real question is..

and the answer is : 42

Do you mind explaining why?
I can imagine a GTO strategy being somewhat stable but not predictable as the frequencies will be mixed in a lot of situations.

This post is for those who fear bots will take over HULHE/HUNL.

I'm an online chess cheater and have been banned from both FICS and ICC so many times I've lost count. I've tried everything, including having the engine on an alternate machine, adjusting the engine so that it chooses subpar moves here and there, making human moves and only using the engine for the most difficult spots yet it has never taken more than a few days for my account to be closed.

Stars/Full Tilt wont have much of a problem replicating this feat. Bots will only be a threat on sites that you shouldn't play on anyway.

There are situations in chess where a strong human player would make an obvious strong (game winning) move, but a computer would make a slightly stronger, also game winning, but counterintuitive move a human grandmaster wouldn't even consider.

The detection of a typical computermove in poker should be similar, albeit i agree with you it wouldn't be trivial to detect due to the smaller number of possible significantly different moves in poker (simply because i assume such situations would come up less often than in chess).

I agree with yuk though that a good exploitative bot would be harder to detect because its moves would look more human (also an assumption ldo).

How about a bot telling a human what to do and the human just executes the actions?

Modern algorithms converge *towards* a solution, they dont simply have a solution either solved or not solved. What he is likely saying is that they solved it to a point that is so close to 0 that it doesnt really matter any further. They wouldnt have "considered" incorrectly, some people may just have a different definition of what should be considered effectively solved.

Chess sites dont care who uses the bot and have software in place to detect if strategy assistance is being used. If your bot had a ****ty strategy, this would likely work, but if you were using any "modern" strategies, then the site would likely detect this fairly quickly

right after that post I looked and wikipedia has a page called solved game. The term has definite meaning(s).

I can see how that detection can work in chess, but how would it work in NLHU?

You can detect bots by looking at consistency of play or multiple accounts playing identically, but detecting "strategic assistance" like the chess sites do doesn't sound viable for poker.

I represent DesertCat Labs and I’m going to answer to some of your questions. Below you will find an answer to the question we’ve received from one of 2+2 member, specifically “We want to know if you have solve the Full Ring limit holdem or not”

The game is solved alchoholicly. We get to the level of precision high enough to consider FLFR solved. There is no problem for us to increase level of precision say by one or more additional rounds of drinks. However it’s not evident if this is practically reasonable because it takes some time plus there are some more interesting problems we have that can be solved by tipping the floor to get free replacement drinks used to solve FLFR, as well as recommendations on cheap hookers.

No approximation or Monte Carlo simulation, as well as CFR was employed during calculation, as typically during our work day we are unable to operate a computer. The algorithm searched every possible drink and cards combination, pocket cards and found optimal action for every possible bartender/server history. For this strategy we didn’t use our supercomputer, other than as a location to leave empties on.

Also I would like to thank LeoDoc for his activity and help with explaining current state of research in poker AI (alcohol influenced) field. Also we really appreciate Alberta Big Rock Brewery contribution to Microbrew Game Theory and team beer related publications.

Unfortunately Beero is a commercial project so we couldn’t go into detailed discussions about our approach and prefer to keep this confidential as much as we can to protect our know-hows and DUI histories. So I will be unable answer any further questions that could ever possible confirm or deny the ludicrous claims I have made.

thanks,
and may I have another,
DesertCat

damn I took that serious well into the second paragraph

Yes, indeed, I'm sure that most of the research my Central/Eastern European postdoc boys have conducted was done within the AI (Alcohol Influenced) field due to the unavoidable cultural specifics.

In fact, being an old man who likes to imagine himself as a patron of the arts and occasional sciences, I made the initial investment as a type of charity as the boys looked bright with a hungry gleam in their eyes. At the time of the initial investment they were both broke and sober.

So it came by surprise that ever since meandering on the elysian fields of alcoholic stupor it seems they did stamble by mistake no doubt into some "algorithm" that seem to be working. Dont drink when you drive, but do so when you programm. - OptedOUT

I have the boldness to guess that the appoach of Alberta lab boys was rather different.

I understand you don't want to go into details about how to solve etc. but giving a number describing how good the solution is (how close to Nash) could only serve as advertisement to you. At least please give the ballpark

No. What he wrote means that they are so close to the solution that he considers it solved and that there is no problem to going closer but he considers it not necessary. There is nothing about possibility of him being wrong.
Those things are easy to verify, if they verified that they are say 0.001bb/100 from Nash with methodology to go to 0.001bb/100 or 0.0001bb/100 at will that would describe what he wrote.