It's probably a misquote. The supercomputer easily costs that much. Tuomas Sandholm specifically said on stream that anyone with their code or skills/papers could spend ~$10k on Amazon AWS/EC2 and run a bot just like Libratus.

Yeah, exactly. We are saying the same thing - it worked from the state on the turn to solve the endgame, that's what took the computation/time, because of the in-game solving.

"We" (regular players) use bb/100, as you know. Computer bot tournaments use milli-big-blinds per game (mbb/g (which reasonably is "1000th of a blind per hand")), which measures the same thing but in a different scale. 220 mbb/g is 22 bb/100

post #329
post #332 (quoting post #329)
post #1204
best to heal your biased mind with some modesty

Its basically common knowledge at this point. Was confirmed on stream by the programmer/developer guy who i cant remember the name of (not tuomas)

so does the bot bluff?

Why would it not bluff? You think bluffing requires an ego?

Pretty sure he forgot the "lol" at the end of the post.

twitch chat trolling is head and shoulders above other popular mediums

permission to come aboard

isnt this super bad for poker? are they gonna bring this online or publish their papers/documents? i rarely say that poker is dying but if this became public it would definitely be pokers last nail in the coffin

This is super bad for HUNL in the same way that Cepheus was bad for HUFL because now there is no way to gain an edge on someone who has studied the solution sufficiently well. Yoghii still gets action from the occasional rec at the highest stakes though.

However there is still no serious progress towards a demonstrably very strong 6Max NL bot, in part because the definition of a solution is harder to pin down in a multiplayer game.

There is also no real incentive for people to develop the 6Max AI now that CMU have been the first with a complex incomplete information game and cracked HUNL. Likewise, PLO is safe for the foreseeable future.

But the presence of solutions doesn't prevent people playing their own strategies against one another, especially when the solutions are so complicated. Chess is still played competitively, even though computers have beaten humans for a long time.

Its def not as bad for hunl as cepheus was for limit holdem. Libratus' strategy (which more closely approaches GTO than what any hu pro is doing now) is too complex for humans to properly mimic and really has just shown us all that there is still way, way more to learn.

The distinction has more to do with betsizing and frequency than anything else. The minbet nature of limit holdem is what makes cepheus' strategy easy to copy/implement. Its really not even in the same world, let alone the same ballpark

I'm not worried about humans trying to copy its strategy and making the game tougher, with all the solving tools out there that will inevitably happen in the long run anyway.

I'm more worried that this might run on a desktop pc one day in the near future and you'll find this bot at your tables. I mean I dont know if they have any intentions in solving SH/FR-games too but if they did and published their code it would be disastrous for online-poker.

These bots aren't anything new. They have been playing and winning online since 2010. They've gotten better and better everyday and Labratus in 2017 is where they are at today. Their profits are in tens if not hundreds of millions by now.

IMO this university and professor taking credit for the work Eastern European coders did years ago. They are the true evil geniuses.

^^rofl

Really?

Some reading for you my friend - Enjoy!

http://forumserver.twoplustwo.com/56...ngame-1112918/

http://forumserver.twoplustwo.com/28...-ring-1074250/

http://forumserver.twoplustwo.com/15...-them-1537778/

http://forumserver.twoplustwo.com/29...oming-1514010/

you arent really comparing a few bots which can beat smallstakes to a bot which literally beats anyone and will do so for all times?

if this bot is really running on a smartphone in 5 years it will be so wide spread that nl2 will be tougher than nl5k+ is today.

It's more or less the evolution of these bots that can solve incomplete information games. You made it sound like this was groundbreaking news to you.

NL holdem is an excellent training ground for programs to solve incomplete information. Because of this, programmers have spent hours and hours perfecting their decision making processes. Bots have been killing the poker ecosystem for years. They sucked in late 2000s, could beat 100bb 600NL+ in 2012 and now able to beat some of the worlds best in 2017.

I wouldn't worry about random Joe Schmoe running an app on his phone... these programs are insanely valuable as they have the potential to help solve a number of other situations that involve incomplete information. This isn't code someone will just happily upload to the world.

Libratus isn't simply an evolution on the pokerbots that have been beating small stakes. Comparing them is like comparing a bicycle to a Saturn V rocket.

It is groundbreaking and this bot doesnt have much in common with the poker-bots you are referring to. The bots you're talking about are most likely scripted, they dont do any solving in-game, they have no real intelligence, they simply execute a few lines of codes. This bot creates an almost perfect solution for every possible scenario in-game. The bots you are talking about would probably require trillions of lines of code to function that way, which would be impossible to create and which is the reason why these bots will never be a real threat. You're comparing hand-calculators with super-computers here, there is a massive difference.


There might be no reason to be worried, as the results of their work might never get published or they might not even care to solve SH/FR games. But somewhere in this thread someone quoted one of the developers who supposedly said that this bot could run on a smartphone in 5 years. IF that happens and if it becomes accessible to "random Joe Schmoe" then I'm pretty certain that this is the end of online-poker. I know thats a lot of IF's but I wouldnt take the situation lightly.

they dont care at all to solve SH/FR games btw, and explicitly stated that they are not going to. poker is not the end game for this AI application

oh nice then, good to know,ty

basically this shouldnt be thought of as "the developers created a bot for HUNL" so much as "the developers created an algorithmic AI machine that is capable of teaching itself the GTO solution to games of imperfect information, of which HUNL poker happens to be the benchmark medium for development rather than the final application of the technology."