"Facebook's new poker AI beats best players of the world...6-max".

"We're not open sourcing it…one reason we’re not is that poker is played commercially and we felt open sourcing could negatively impact the community," Facebook spokesman Ari Entin told Forbes. "

no money in poker, every bot's solid!

interesting reads:

https://www.forbes.com/sites/samshea...s-top-players/

https://www.theverge.com/2019/7/11/2...layer-no-limit

https://science.sciencemag.org/conte...cience.aay2400

Honestly I get it from a marketing perspective, but from a scientific standpoint.... picking midstakes 6-max grinders is probably better than picking guys who specialize in live tournaments.

Open Source CFR+ (University of Alberta - HU Limit Hold'em)

Play against Cepheus (University of Alberta - HU Limit Hold'em)

The scary part:

Flip side of that is it appears to of been one match. The worst poker player in the world can win one session with some run good.

10k hands isn't much. Interested in what Linus has to say about it.

https://ai.facebook.com/blog/pluribu...-player-poker/

Video showing hand samples is fun

they used a variance reduction technique that supposedly reduces the # of hands needed to be considered a reasonable sample by a factor of ten or more

How did Ferguson end up playing a 6max cash game bot for a project like this in 2019?





Looks like Linus was in the mix for this as well. He almost broke even playing vs 5 of the bots

Sandholm and Ferguson have been close for a long time. He invited Ferguson to CMU to consult on earlier poker AI projects back in like 2007.

Yeah, this is weird. I don't know much about Elias but Ferguson always seemed like a) he played very ABC, and b) he was one of the tighter players, which would not lend well to short-handed play.

Probably wanted an opponent that wouldn't give him a constant barrage of snide remarks. Besides, his apology video was pretty robotic -- maybe he felt at home in the Carnegie Mellon computer lab.

This said and kidding aside, it's still pretty impressive. If I built a robot that raked in a month's worth of games in the Double-A Eastern League, I would hope the first response isn't "meh, call me when it can hit off Verlander and Scherzer for a full season."

EDIT: Never mind, saw the full list of pros in that Facebook article – Jimmy Chou, Seth Davies, Michael Gagliano, Anthony Gregg, Dong Kim, Jason Les, Linus Loeliger, Daniel McAulay, Nick Petrangelo, Sean Ruane, Trevor Savage, and Jake Toole.

Definitely, some unique lines being taken...


SB BB UTG HJ CO BTN
Gogo Budd Eddie Bill Pluribus MrWhite
Jh9c Ts6s 2d7s Kh9s Kd7d AdJc
-50 -225 0 0 500 -225


PF Action:ff r225 c f c
F: 4s5d8h
F Action: c r1675 f f

I'm sure the source code has already been hacked, or will be, and leaked to the dark web as we speak

I must be unable to find the link to the HH's

Is it in the FB article??

https://science.sciencemag.org/conte...ce.aay2400.DC1

According to at least one media group the AI has already been shared with other researchers in other Universities/groups.

Considering that some of these researchers are PHD/ postgraduate students it may as well be open source.

Clearly Carnegie-Mellon has no issues with releasing an AI that can used for deceiving poker players.

Can't comment on the bot itself too much because all the media reports are pretty different.

Most of them don't even confirm if bot was playing an SNG or Cash (sounded like SNG to me)

O/U the # of years it will be until online poker is about who has the most advanced bot?

Ill say 15 years

2011- near-apocalypse for online poker - Ferguson running Full Tilt at the time
2019- Jesus Ferguson colloborates with the top AI researchers in the US to create unbeatable bot thus facilitating another poker apocalypse
2026- Chris aids in development of sentient bot that destroys all humans in every poker variant


TLDR : Jesus is the anti-christ of poker

After 40-second-tank Libratus beat humans they said even 3-way is still beyond far from being tackled by AI. I was pretty happy about that but now we have lightweight bot breaking even vs Linus at 6m. Lol. Sad times for online poker.

https://arstechnica.com/science/2019...-texas-holdem/

pretty sick. can someone convert these to be able to import into pt4?

and is there one for libratus hu match?

I'd prefer if the lineup included the stars 25/50nl 6m regs and a bigger sample size.

This is all very depressing nevertheless

https://www.theverge.com/2019/7/11/2...layer-no-limit



https://science.sciencemag.org/conte...cience.aay2400


https://ai.facebook.com/blog/pluribu...-player-poker/





https://www.facebook.com/FacebookAI/...7259478050682/

I think these details are whats scariest about this development.

you should try to get linus to talk about it on the podcast.

also i wonder if this isnt as bad for online poker as it seems because if its not costly to run and it becomes open sourced someday maybe this will also create advances in bot detection for poker sites.

also 10k hands isnt much , so run good is still a factor , wish the would have played more hands.