Some chess bots beat other chess bots.


I reckon a human will rule poker forever, becaUse GTO is inherently flawed. When it comes to timing,//tempo.

Has anyone been able to develop something custom that reads data straight from an active table an initiates a simulation? If not, that really seems like the next frontier of cheating capabilities, no?

If yes, presumably they'd keep that pretty close to the vest.

A cheater probably doesn't even need custom solutions for each spot. You can develop some sort of model that buckets similar textures and lines and just spits out some reasonable approximation. Surely with such a model you could quantify ex post how similar/dissimilar each output is and refine the model accordingly.


Thought experiment: if you were going to build a predictive model for identifying a cheater, what would be the most important variables to capture? I'm thinking
-Winrate
-Volatility (standard deviation of winrate) (honestly not sure if there's any predictive power with this one)
-Avg number of tables played at a time (if you still have to manually look up solutions you'd be pretty time crunched if playing a lot of tables)
-Anything to glean from average time to act? Either faster (because it's a computer doing the work) or slower (because they're constantly looking up solutions)
-What else?

Feel like comparing a credible sample of actions to Nash solutions and computing exploitability is not gonna work because it doesn't capture the intention to exploit so a cheater may very well be deviating quite a bit on purpose, as any competent person should wont to do it they think it's justifiable.

It's a two-sided thing: you'd want to know your own strategy but also opponent's response. So I feel like any anti-cheating model needs to incorporate population data to flag where people (assuming you've got data from a representative sample of non-cheaters) are most exploitable and specifically focus a lot on the potential cheater's behavior in those spots. But then you have this big issue of false positives: honest people may rightfully get pissed if they're penalized for being a sicko.

What about subjecting a potential cheater to some sort of monitored test? If they're flagged, they get randomly assigned a sufficiently large number of sample spots and, under timed supervision, they have to fully specify all ranges, strategies, and justify any deviations they'd make from Nash on the basis of their assumptions about opponents' behavior. If you can't actually do these computations from "first principles" but you're crushing then there might be a problem. Force people to "show your work".

I haven't really contemplated the effects of technology improvement on this stuff till now and tbh this reality sucks. This is a tough issue to tackle. And I feel like it should be taken as a given that nefarious people will make every attempt to exploit this.

Am I being too much of a tin foil hat conspiracy guy to think that there may be a shitton of people already using advanced cheating methods? Do online sites really have ANY mechanisms in place to stop this? Do they even have any incentive to stop it yet?

Why not just let every single online poker player use an RTA standardized GTO bot with a "RTA" inbuilt and let 'em all fight?

A potential simple solution (?):

Mandatory screen recording software for all active monitors. And mandatory webcam. And lock down access to all web browsers and desktop solvers while the client is open?


Is this just too invasive to ever be an acceptable solution? Plus would sites have to invest in better infrastructure to process all the data?

If everyone is using a model that Is equal it's fair game. If one is using a quantum generator of node/divergent//tree-route logic that goes helluva deeper then the competitor, then that's a huge problem. Think IBM deep blue versus chess computers now

You edited your post.

Screen recording is interesting. I think it poses more problems than solutions at current;;

Time delays, deep fakes, and collusion

Invasiveness is another issue. Lots of sites have tried and failed.

The average winning non-cheating player seems to base most of his game on heuristics based on things like formation, action, texture etc, and the improvements come from expanding and refining those heuristics. Add MDA for some river exploits and you have it. If a bot can just be like that, and execute a very good human strategy without ever making a mistake (inside the human strategy) or tilting, I think he will be winner. Not perfect at all, but perfect is the enemy of the good (good=stealth).

Is everything I said above implementable by any decent programmer? Absolutely.

The infrastructure to execute this would be the main problem from my pov.

EggsMcBluffin:

- Winrate first.
- If all of this end up being deployed as bots, I assume it would be deployed with tons of accounts playing the same way, so I guess a site would have to constantly watch for players playing the same.
- Someone playing big volume, picking random 100k, 200k samples and scrutinizing the stats just like you would do to catch a full botfarm, as I don't think any human would be able to execute the exact same strategy he did in 100k hands, over the next 100k hands. Including showing the same leaks.
- + anything else others said or may say itt

Bloom; first paragraphs true

As for;;;

- how do some bots make more money deviating from GTO for exploits from data affect winning players ~potential win rates, IF that's in their programming

- no bots, fish whales and sharks and shrimp
- all bots, the well dries up


Also, worth to think;;

If bot farms are targeted and one true player gets banned, is that worse than allowing break even slightly ++ bots from winning????

Any bot running at even close to 70%+ optimal should be sniffed out ez

I believe fish gonna either lose less from bots or worst case the same than vs regs. Bots are bad for regs in my opinion, and most of the complaining is just self interest.

Unfortunately, lots of fish fear bots, rta and cheating, they fear or used to fear huds lol. With no fish, regs will slowly disappear, and no players=no games.

My 2 cents

They play vs fish way better than the average reg

Live poker boom incoming.


My 2cents online poker??

Old ''regs'' will become low losing recreational


Some bots will eak out ~2bb/100

Some meta-sharks will have 5bb win rate

Okay... Maybe bots win 6bb/100

Sharks [human] win at 2.



The online poker dream is dead courtesy of online strategy vids, theorems, and tech

My bad, fish has very predictable leaks that any bot programmed with mda data and the adjustments a good reg would make will just destroy. It’s gonna be a massacre lol

A bot would never 2nd guess a close bluff catch for stacks vs a fish after losing 10 times in a row because he thinks that fish is different than the average fish or whatever.

Nice attempt to patronize. I'll give you a hint and I'll leave it at that. The unit you measure std deviation in poker is bb/100. Variance in shortstacking poker (and mtts) is lower in terms of bb/100 but higher relatively to your stack. You will get stacked more often since you are all in more often which is good for the speed of the tournament. But the swings and the volatility of them will be far lower the shorter you are. High risk = high reward , small risk = small reward.

The dream of being able to roll out of bed stumble to your computer and profit is certainly dead. You can still make money if you put in time and effort