|
|
Poker bot "Lengpudashi" dominated 6 Chinese players in Heads-Up matches |
|
5
|
Garfed   Malta. Apr 13 2017 18:35. Posts 4818 | | |
The updated version of Carnegie Mellon University heads-up no-limit hold’em poker bot (which you should remember under the name Libratus) easily defeated a team of six well-known Chinese players last weekened.
The match last five days and ended on Monday, and the report from Bloomberg stated that the contest was a “landslide” for the bot and it was a one-sided game. Venture Capitalist Alan “Yue” Du was the head of Team Dragon, who competed against the bot. Du made history in 2016 when he defeated a field of 863 players in the $5,000 No-Limit Hold’em World Series of Poker (WSOP) event thus becoming the first man from China to win a bracelet.
The machine, called now "Lengpudashi", won the $290,000 prize for its creators.
The game in Chineese provine in Hainan lasted just 36,000 hands. The smaller sample size meant that the contest wasn’t a “scientific experiment” like the battle earlier this year, according to CMU.
“This is an exhibition, not a match, challenge or competition,” computer scientist and bot developer Tuomas Sandholm said. “We are running a relatively small number of hands, so this is not a scientific experiment.”
|
|
| 4
|
Baalim   Mexico. Apr 13 2017 22:56. Posts 34246 | | |
36k hand isnt determining at all, but I think its clear that HU AI is very close to top humans, damn. |
|
Ex-PokerStars Team Pro Online | |
|
| 1
|
traxamillion   United States. Apr 14 2017 01:36. Posts 10468 | | |
It's significantly better than any human.
This version wasn't even run on a supercomputer. It didn't have the massive power to run the turn/river endgame solver that it had versus Les Donger Swarm and Dougiedan. It still wrecked though.
If you watch the video of it playing the Asians though you can tell they were pretty awful at HUNL |
|
| Last edit: 14/04/2017 01:38 |
|
| 1
|
Highcard   Canada. Apr 14 2017 05:03. Posts 5428 | | |
|
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 | |
|
| 4
|
Baalim   Mexico. Apr 14 2017 07:22. Posts 34246 | | |
| On April 14 2017 00:36 traxamillion wrote:
It's significantly better than any human.
This version wasn't even run on a supercomputer. It didn't have the massive power to run the turn/river endgame solver that it had versus Les Donger Swarm and Dougiedan. It still wrecked though.
If you watch the video of it playing the Asians though you can tell they were pretty awful at HUNL |
It is possible but there is no way you can say with any remote certainty yet |
|
Ex-PokerStars Team Pro Online | |
|
| 1
|
lucky331   . Apr 14 2017 17:37. Posts 1124 | | |
chong choi la! chi cha twong eee err errr. kekekeke! |
|
| 1 | |
from results so far it seems very likely that recent versions of this bot is better than any human at hunl. To my knowledge it hasn't lost against anyone and varied between crushing and merely defeating everyone? If I'm wrong about that, if it lost against anyone, then my apologies. And yeah sample sizes aren't conclusive, obviously this one isn't, but the previous sample size was 120k hands.
I'm curious and I dunno how to calculate this but there are some math wizards here. What's the likelihood of a non-winning player going 14.6/100 over 120k hands? Or, looking strictly at vs dong kim who did the best, like 2.8/100 over 30k? |
|
|
| 1
|
traxamillion   United States. Apr 14 2017 22:45. Posts 10468 | | |
It hasn't lost to anyone. Dong also had the better run of cards, his mirror got crushed. The bot smashed some top pros with scientific certainty. Sure there might be less than ten better players, but how much better? This bot is insanely good |
|
| |
|
|
Poker Streams | |
|