Perfectly legitimate question. Poker is actually my emergency back-up career - I did 25 years in the software engineering industry including stints at IBM, Silicon Graphics, and Synopsys before joining PokerStars in 2003.

I actually see working in the poker industry as providing entertainment (via poker) to the world; I believe it's an honorable pursuit. And I've been fortunate enough to make some money that I can hand over to people smarter than I to address things such as climate change and civil liberties.

But I don't want to hijack the thread (it's the most interesting thing I've read in NVG in a long while); feel free to DM me with further thoughts.

Best regards,
Lee Jones

Thanks for the response.

If this AI can be adapted to any game, so theoretical war-game scenarios, and find better GTO solutions than the current top minds, how is the military not interested? Or are they?

Oh, then you can disregard my posts as I hadn't watched any of the streams but was going by what was being said itt which seemed to be suggesting tank times in the 2 minute region. If Lib is able to fully solve turn/riv given its assumptions of going-in ranges within 30 seconds on average then yeah, gotta say fair play and gg humanity :/

the debilitating part of it is that it tanks 15-45 seconds in very tiny pots, so it will take 45 seconds to check turn OOP, hero bets, and then itll take 30 seconds before it decides to fold turn. pretty tilting in a pot of 500 or w/e

^That just seems kind of buggy... have to assume it's been thrown out anew by the bet sizing, but surely in a spot like that it should have all prospective sizes bucketed such that the initial tank has everything covered already.

ummmmm

far less sophisticated bots have won more than this, and that's likely just the tip of the iceberg. what are you even talking about?

Is there any current bots that use decision trees and brute force to play? As far as I know all the decent bots are just preprogrammed with a static range pre/post with option to deviate vs specific screen names or vpip/pfr %. None really can figure out optimal strategies themselves.

oh, idk, i posted the question because i was not convinced that the mirroring would reduce variance by all that much. if we assume that a perfect strategy involves a mixed strategy for a lot of hands with different betsizes and all, i would imagine that even if 4 players knew GTO and would play in a mirrored hands setup, there would still be a lot of room for variance. imagine for a specific hand in a specific spot it was GTO to bet flop for x1$ y1% of times and z2$ z2% of times (...) and check z3% of times and so on, outcomes in the mirrored hands could vary greatly. but idk, my understanding of math is too bad to really grasp this.

What's the link to the stream?

Kind of a derail, but the military already has links with the CMU Robotics Institute. https://www.ri.cmu.edu/news_view.htm...74&menu_id=238
I think I heard Uber (which isn't military, obv, but has a huge investment in AI) also had some sort of partnership with CMU and basically poached a load of the staff from the Robotics center.
I cynically see this poker bot thing as more of a 'publicity stunt' that helps attract more research funding (and students) to the computer science dept. It's as yet unproven that shoving 10x pot will cure cancer or bring about world peace.

Libratus in charge of nukes? Would be interesting. I bet he'd find the GTO solution for world peace.

wats the score

baby steps

Good point. Shortly after I posted I actually started wondering what % of deposits are made by players who wouldn't even care if they knew that their opponents were using artificial intelligence tools. I've played against plenty of rich dudes who were smart enough to know they were significant losers in the game, but it was still fun for them and worth a few $k every weekend. Maybe the fun doesn't diminish that much when they know they're playing against AI tools.

I don't know how much Stars should even care about combating these tools/bots. Isn't the dream to have everyone playing GTO against each other?

Extremely unlikely there is any genuine self taught AI in online poker. The extent of online bots is most likely the "if X do Y" type.

looks like the humans improved to -9.7bb/100 hands through 52k hands

The difficulty in "theoretical war-game scenarios" isn't necessarily that they're hard to solve, it's that they're very hard to define. Lots of assumptions have to be made that could have disastrous consequences if they're off by even a little bit. War is also a prisoner's dilemma type game that requires cooperation to avoid mutually assured destruction, so the GTO solution isn't likely to be the best solution.

what do you think the odds are that they use the phrase "statistical tie" if they win by 9bb?

Pretty unlikely, but they are doing 50% more hands this time around so a lower winrate will be more significant.

Stuck over 640k now, this is so sick, curious about nsd

This CMU guy made a page that just live crops the scoreboard from each table so you can sweat the full results during the day without opening all the streams: http://www.cs.cmu.edu/~ecavallo/brainsvsai/

^^ guess they are better at programming soul destroying poker playing robots than a basic web-page to view live images in a tabular format ^^

I'm curious cheet, when you play the bot is it disorienting? ie: in HUNL you can often start to get an understanding of what the opponent is trying to do, tendencies, etc. and try to counter that. It would seem with Libratus you'll encounter strategy that feels random and uncharacteristic and therefore very hard to confidently defend. This may be an obvious thing considering it's a robot and and is designed to balance itself randomly within the parameters of GTO (my interpretation at least). But I'd be interested to hear your view on it?