|
![](/images/design/headeropen.gif) |
Are AI's better than humaans at poker now? - Page 2 |
![](/images/design/headerclose.gif) |
1
![](/images/icons/1/9.gif) |
RiKD   United States. Jun 09 2024 04:48. Posts 8823 | | |
Pluribus was a loser in the sample (so was Linus) but they used a paper on variance to prove an EV of 5bb/100. I didn't read the paper. One thing I found interesting about Pluribus is it didn't overbet. Which obviously overbetting is a large strategy of the solver. I think the biggest risk is RTA (Real Time Assistance) using solver strategies. Or, a solver bot that includes overbetting in their strategy. I don't see why Pluribus had like multiple 3 bet sizing but not a myriad of post-flop bet sizes. 1/5, 1/4, 1/3, 1/2, 2/3, 3/4, 4/5, 125%, 155%, 200%, 300% ... All-in. It doesn't get any bigger than Facebook AI and Carnegie Mellon University so maybe this is actually really difficult to do. But, I would bet on the Russians doing it. |
|
| 1
![](/images/icons/1/11.gif) |
Stroggoz   New Zealand. Jun 12 2024 22:51. Posts 5323 | | |
Pluribus is solving for 6max entirely, whereas pio is just customized 2 way scenario's given certain decisions in a 6max game. I'm pretty sure Pluribus lost out a decent amount of ev in not having more sizes, but given it's actually training for the whole game and not customized scenario's, It has to simplify to account for the added time complexity. |
|
One of 3 non decent human beings on a site of 5 people with between 2-3 decent human beings | Last edit: 12/06/2024 22:52 |
|
| 1
![](/images/icons/1/9.gif) |
RiKD   United States. Jun 15 2024 21:44. Posts 8823 | | |
I saw the creaters of Pluribus were bragging about the computer usage (reduced time complexity?). I want to see what happens if you don't care about that stuff and give it a full range of bet sizes.
I saw an article by Peter Clarke where he said a solver wouldn't win at 25 NL. Too much rake and the exploitive players would be winning all the money from the recs. Yes, the exploitive regulars would win more from recs (probably?) but the solver bot would be winning from the recs and the regulars. Recs are losing at -2 to -8+ bb/100. Bad regs are -3bb/100 or so. Good regs are 3-4+ bb/100. Rake is 8 bb/100. In that scenario I think a solver would have to be winning probably 5+ bb/100. |
|
| 1
![](/images/icons/1/11.gif) |
Stroggoz   New Zealand. Jun 18 2024 07:49. Posts 5323 | | |
Likely there will never be a computer or algorithm ever found that shows you how optimal poker would work given a full range of bet sizes. It would use too much compute.
I wouldn't be suprised if Peter Clarke is right. 25NL skill level is probably like 4-10bb/100 difference in skill from someone like LinusLove. The gap has closed considerably. |
|
One of 3 non decent human beings on a site of 5 people with between 2-3 decent human beings | |
|
| 1
![](/images/icons/1/9.gif) |
RiKD   United States. Jun 20 2024 06:16. Posts 8823 | | |
I remember you saying that now a days there are a lot of people just as good as Linus if not better but that Linus gets the credit for being an early adopter.
So, there is a rush of these solver guys who can also be exploitable and the recs are getting eaten alive. I know ACR lobby looks terrible most of the time and some 10 NL lineups today are tougher than hell, some 5,000 NL lineups in 2009. |
|
| Last edit: 21/06/2024 03:40 |
|
| 1
![](/images/icons/1/9.gif) |
RiKD   United States. Jun 23 2024 00:07. Posts 8823 | | |
Ya know, I let the above post sit for a bit and I think I am grasping here. A 6-max table of the 6 best 10 NL players on ACR are probably better than a 5,000 NL table with a bum and 5 bumhunters in 2009. That isn't really a fair comparison though and it might not even be true. Today's players will have a much better pre-flop game and probably flop game but I am not so certain that they would play turns and rivers better. Bet-sizing has come along way though too as well as a lot of concepts. I don't know. It's an interesting thought regarding just how fast the games progressed in 15 years. |
|
| 1
![](/images/icons/1/9.gif) |
DooMeR   United States. Jun 25 2024 00:38. Posts 8553 | | |
| On June 22 2024 23:07 RiKD wrote:
Ya know, I let the above post sit for a bit and I think I am grasping here. A 6-max table of the 6 best 10 NL players on ACR are probably better than a 5,000 NL table with a bum and 5 bumhunters in 2009. That isn't really a fair comparison though and it might not even be true. Today's players will have a much better pre-flop game and probably flop game but I am not so certain that they would play turns and rivers better. Bet-sizing has come along way though too as well as a lot of concepts. I don't know. It's an interesting thought regarding just how fast the games progressed in 15 years. |
i wouldnt say 10NL but definitely 200NL for sure is way tougher than any 25/50 back in the day. Maybe 100NL on sites like pokerstars but not rly sure. U transplant durrr from 2009 he is gonna be a losing player at 1/2 blitz for sure. Rofl Id go so far as the durrr from today would be also, but thats his fault no one elses lol |
|
I just saved a bunch of money on my car insurance, by running away from the scene of an accident. | |
|
| 1
![](/images/icons/1/11.gif) |
lostaccount   Canada. Aug 12 2024 04:42. Posts 6089 | | |
|
humble and modest living, my karma is done, now time to enjoy life, peace is the way karma is a way Jesus is a way. peace of mind is the greatest gift liquidpokerIG | |
|
| |
|
|
![](/images/design/space.gif) Poker Streams | |
|