5
|
Garfed   Malta. Jan 24 2017 22:25. Posts 4818 | | |
The 2017 version of Carnegie Mellon University’s poker pro vs. poker bot contest hit the half-way mark yesterday morning and the team represented by real humans is falling behind faster then ever! Check the results at the end of this news!
With 64,000 hands of the 120,000 scheduled played the poker pro team of Jason Les, Dong Kim, Jimmy Chou and Daniel McAulay is down nearly 800,000 chips. That’s 8,000 big blinds. None of the pros is ahead of the machine. The contest is scheduled to end Jan. 30.
Early last week, the humans were showing signs of little comeback after they started breaking even in games, but the machine was having none of that and after some adjustments to the strategy it started to completely destroy the players:
“The bot gets better and better every day,” Chou told CMU. “It’s like a tougher version of us. The first couple of days, we had high hopes. But every time we find a weakness, it learns from us and the weakness disappears the next day.”
Libratus was developed by Tuomas Sandholm, professor of computer science at CMU, and his Ph.D. student, Noam Brown. Libratus is just the latest in a series of poker bots from Sandholm.
According to comments from the human team, this version of the bot adjusted its river play according to previous machines and Libratus will often over-bet shove on the river for either value and as bluffs. While humans show some bias in such a river strategy, the machine has remained balanced and thus pretty unpredictable.
Each night after a session, the Pittsburgh Supercomputing Center’s Bridges computer performs computations to sharpen Libratus’ strategy, especially with regards to the river.
Here are the results:
Each battle is live streamed on Twitch, you can find them under those links:
https://www.twitch.tv/libratus_vs_jasonles
https://www.twitch.tv/libratus_vs_dongkim
https://www.twitch.tv/libratus_vs_jimmychou
https://www.twitch.tv/libratus_vs_danielmcaulay
|
|
| Last edit: 24/01/2017 23:05 |
|
|
1
|
Stroggoz   New Zealand. Jan 24 2017 23:27. Posts 5296 | | |
i'd kill for some HH's from this bot |
|
One of 3 non decent human beings on a site of 5 people with between 2-3 decent human beings | |
|
|
1
|
spets1   Australia. Jan 25 2017 06:49. Posts 2179 | | |
make it real money and real pros then see how it goes.
are those guys considered any good? What stakes are they winners at? |
|
|
|
1
|
Trav94   Canada. Jan 25 2017 07:29. Posts 1785 | | |
| On January 25 2017 05:49 spets1 wrote:
make it real money and real pros then see how it goes.
are those guys considered any good? What stakes are they winners at? |
As far as I know, all of them except Dan Macauley play 50/100. Macauley might too, I just have no idea who he really is. |
|
|
1
|
Expiate   Bulgaria. Jan 25 2017 07:59. Posts 236 | | |
| On January 25 2017 05:49 spets1 wrote:
make it real money |
Maybe this is coming next year, but who will want to play then? |
|
|
1
|
spets1   Australia. Jan 25 2017 08:13. Posts 2179 | | |
players like isildur, smash some cpu for money? |
|
|
|
1
|
Stroggoz   New Zealand. Jan 25 2017 10:12. Posts 5296 | | |
i'd imagine isildur would be alot weaker than these guys having been on break from nlhu for a long time. I used to play Dan hu at .5/1 5ish years ago, and i saw him at 50/100 2 years ago so he moved up pretty quick. |
|
One of 3 non decent human beings on a site of 5 people with between 2-3 decent human beings | |
|
|
1
|
Spitfiree   Bulgaria. Jan 25 2017 10:35. Posts 9634 | | |
Cmon dude, its the 3rd or 4th time in a row you post news that are 24hrs old. They won 10bi yesterday, down 70bi in 71k hands resulting in around 10bb/100 loss
https://www.twitch.tv/libratus_extra
There you can see the info up to date.
The guys are complaining about the bot adjusting instantly, when they adjust to him and that they have no means of exploiting him, while he destroys them. |
|
|
1
|
Stroggoz   New Zealand. Jan 25 2017 12:36. Posts 5296 | | |
Donger is actually beating the bot, by four buy ins, very close. Seems like if it was just 4 Dongers or people better than Donger then the humans could have a chance vs this bot still. Jason Les and Donger get dealt the same hands vs the bot right? |
|
One of 3 non decent human beings on a site of 5 people with between 2-3 decent human beings | |
|
|
1
|
lucky331   . Jan 25 2017 13:52. Posts 1124 | | |
| On January 25 2017 07:13 spets1 wrote:
players like isildur, smash some cpu for money? |
i was thinking he same. blom could break that bot. |
|
|
1
|
Expiate   Bulgaria. Jan 25 2017 14:44. Posts 236 | | |
I highly doubt it, the improvement of the bot from the last year is tremendous. Blom could be ahead, but no way to crush it for say more than 20 bb/100. |
|
|
1
|
punix   Germany. Jan 25 2017 15:10. Posts 406 | | |
would be interesting if the bot would win in 6handed games |
|
|
1
|
cariadon   Estonia. Jan 25 2017 17:42. Posts 4019 | | |
lots of talking out of the ass |
|
|
1
|
ToT)MidiaN(   United Kingdom. Jan 25 2017 20:10. Posts 5070 | | |
| On January 25 2017 11:36 Stroggoz wrote:
Donger is actually beating the bot, by four buy ins, very close. Seems like if it was just 4 Dongers or people better than Donger then the humans could have a chance vs this bot still. Jason Les and Donger get dealt the same hands vs the bot right? |
According to https://www.riverscasino.com/pittsburgh/BrainsVsAI/ Donger Kim is up 12.6 buyins in fact, unsure if this is up to date? |
|
One day good. One day bad. And some days, even hope | |
|
|
1
|
FullBRing   Philippines. Jan 25 2017 20:11. Posts 581 | | |
Wouldnt change anything for the bot to play real money, would only if anything, play against humans ^^ |
|
|
1
|
Spitfiree   Bulgaria. Jan 25 2017 21:02. Posts 9634 | | |
| On January 25 2017 19:10 ToT)MidiaN( wrote:
Show nested quote +
On January 25 2017 11:36 Stroggoz wrote:
Donger is actually beating the bot, by four buy ins, very close. Seems like if it was just 4 Dongers or people better than Donger then the humans could have a chance vs this bot still. Jason Les and Donger get dealt the same hands vs the bot right? |
According to https://www.riverscasino.com/pittsburgh/BrainsVsAI/ Donger Kim is up 12.6 buyins in fact, unsure if this is up to date?
|
He was ahead around 4bi yesterday, its 12.6 now yes. Twitch page has the up-to-date results at all times |
|
| Last edit: 25/01/2017 21:02 |
|
|
1
|
Highcard   Canada. Jan 25 2017 21:09. Posts 5428 | | |
Ok Dan is dougiedan from stars, he plays the highest HU now after daily grind.
Jason les, like last time, is the fish. Delete him and they would have a much better chance.
They all get the full hand logs with the holecards at the end of the day. Those are private. These 4 players are going to have super strong knowledge of how this GTO bot plays and nobody else (except doug, I expect dong and doug talk and other private groups). I guess it is possible they signed a NDA but who knows, the bot creator said the players asked the HH not be released.
After it is all done, the bot creator is doing a massive QA and will explain a lot |
|
I have learned from poker that being at the table is not a grind, the grind is living and poker is how I pass the time | Last edit: 25/01/2017 21:09 |
|
|
1
|
TheTrees   United States. Jan 25 2017 21:21. Posts 1592 | | |
|
|
1
|
Highcard   Canada. Jan 25 2017 21:34. Posts 5428 | | |
just now Jason said he is done $500k. He makes up close to 60% of the losses. |
|
I have learned from poker that being at the table is not a grind, the grind is living and poker is how I pass the time | |
|
|
1
|
Endo   United States. Jan 25 2017 23:10. Posts 953 | | |
Les is just getting raped |
|
|
1
|
Highcard   Canada. Jan 26 2017 00:00. Posts 5428 | | |
damn jason is down 600k now, if he open folds he would do better |
|
I have learned from poker that being at the table is not a grind, the grind is living and poker is how I pass the time | Last edit: 26/01/2017 00:00 |
|
|
1
|
Highcard   Canada. Jan 26 2017 00:35. Posts 5428 | | |
there is a guy in chat namdnasrm who said the bot has no preflop/flop solver and wants them to start shoving preflop/flops to about 80-90%. Saying the bot will solve to the nearest solve that he knows and it will map wrong ranges |
|
I have learned from poker that being at the table is not a grind, the grind is living and poker is how I pass the time | Last edit: 26/01/2017 00:36 |
|
|
1
| |
I want to see libratus playing libratus |
|
Valor pleases you, Crom... so grant me one request. Grant me revenge! And if you do not listen, then to HELL with you! | |
|
|
4
|
Baalim   Mexico. Jan 26 2017 04:50. Posts 34250 | | |
| On January 25 2017 23:35 Highcard wrote:
there is a guy in chat namdnasrm who said the bot has no preflop/flop solver and wants them to start shoving preflop/flops to about 80-90%. Saying the bot will solve to the nearest solve that he knows and it will map wrong ranges |
Now this is how you exploit a bot, they exploited the last bot with overbets, trying to play bots straight is stupid, the point is to find the chink in the armor.
And Jason Les did also horrible the last time why was he invited again? |
|
Ex-PokerStars Team Pro Online | |
|
|
1
|
Highcard   Canada. Jan 26 2017 18:29. Posts 5428 | | |
well from the University they want profish so they can say they beat pros. Easy to invite him again. He said yes obviously because it is the easiest money he can make. If I was the other people I would have voted him off before match started. Could be 1 of many reasons why Doug said no.
The entire point of this is to beat pro humans 1x then they can advertise their university program to phd candidates etc. etc. and get better students/funding.
I read that Deep blue actually lost to Kasparov 1/2 but IBM would never play best of 2/3 because all they wanted was that advertising dollar. It should always have been best 2/3.
If the humans lose it was already said the bot will never play humans again unless a random event happens.
In both instance, these companies have Massive profit ratios vs risk/expense. The humans are being severely underpaid
|
|
I have learned from poker that being at the table is not a grind, the grind is living and poker is how I pass the time | |
|
|
1
|
traxamillion   United States. Jan 26 2017 18:33. Posts 10468 | | |
I thought a good way to try to exploit the bot would be to find it's abstraction threshholds since it essentially uses rounding. If
85% of pot is the same thing to the bot and 109% of the pot is the same thing to the bot as pot (rough numbers obv). You can find an exploit on the bot by bluffing 85% and vbetting 109%.
Attacking the abstraction is key from what i have read |
|
| Last edit: 26/01/2017 18:38 |
|
|
1
|
Highcard   Canada. Jan 26 2017 19:03. Posts 5428 | | |
LOL Jason is on life tilt
an article says he sucks and is playing the bot as if it was a human. Jason tilt chats. Then he starts comparing himself to doug polk saying when doug played 5/10 people said he sucked and would never make, so, ya, he is going to be good too. |
|
I have learned from poker that being at the table is not a grind, the grind is living and poker is how I pass the time | |
|
|
1
|
Highcard   Canada. Jan 26 2017 19:25. Posts 5428 | | |
2 weeks in and now jimmy is giving jason game theory to stop him from endless whines |
|
I have learned from poker that being at the table is not a grind, the grind is living and poker is how I pass the time | |
|
|
1
|
Highcard   Canada. Jan 26 2017 19:47. Posts 5428 | | |
| On January 26 2017 17:33 traxamillion wrote:
I thought a good way to try to exploit the bot would be to find it's abstraction threshholds since it essentially uses rounding. If
85% of pot is the same thing to the bot and 109% of the pot is the same thing to the bot as pot (rough numbers obv). You can find an exploit on the bot by bluffing 85% and vbetting 109%.
Attacking the abstraction is key from what i have read |
yes their open sizes are all numbers they found. |
|
I have learned from poker that being at the table is not a grind, the grind is living and poker is how I pass the time | |
|
|
1
|
Stim_Abuser   United States. Jan 28 2017 08:05. Posts 7499 | | |
|
Hey Im slinging mad volume and fat stackin benjies I dont got time for spellin n shit - skinny pete | Last edit: 28/01/2017 08:07 |
|
|
|