Can AI Spot Poker Leaks That You Don't Even Know You Have?

O

optimus.poker

Enthusiast
Silver Level
ES
Poker Chips
85
Casino Coins
0
Hey everyone! 👋

I've been thinking about something interesting when it comes to improving at poker.

Most of us probably have certain habits that we don't even realise are costing us money.

Maybe we're defending the big blind too often, folding too much against aggression, or consistently making the same mistakes on the river.

But what if AI could identify those patterns before we even noticed them ourselves?

Instead of analysing individual hands, imagine having a tool that looks at your overall playing tendencies and tells you:

  • You're losing too much EV in certain positions.
  • You're consistently overfolding against specific bet sizes.
  • You're making the same mistakes in similar situations.
I think this could completely change the way players study poker, especially those who don't have hours to spend reviewing hands.

I'm curious about two things:

  1. Would you trust AI to identify weaknesses in your game?
  2. Would you prefer an AI that tells you what you're doing wrong or one that actually teaches you how to fix it?
 
I prefer a human being as a pokercoach.
But yes if you feed AI with a big enough hands sample it surely can detect flaws
 
I'd be happy with whatever works, and or whatever is most accessible.

I would trust AI to identify leaks for sure, as long as it had some explanations of the why change, plus benefits etc, alongside the 'leaks' identified.

So yeah, there's little point in something that simply identifies issues if they are not also suggesting fixes and how to implement them
 
It would be interesting to see if AI could win a poker tournament with the cards it is dealt, much like a machine beating a chessmaster at chess.
 
No I wouldn't trust AI. AI only knows perfect poker math and GTO. The list of what AI doesn't know far out weights the list of what it knows.

AI doesn't know player pool tendances, doesn't know exploitive play, doesn't have instincts, doesn't have player reads and doesn't know when or how to bluff. Without having/knowing all them things AI can't tell the difference between a leak, an exploitive play, a bluff or instincts.

Now if one hand a million+ hands to feed AI and knew how to train AI, and knew how to ask AI the right questions because AI can be straight up be wrong and be mislead depending on the way/wording of ones questions seen it many times.......I'd say maybe.....

Even then I'd still pay a coach to review my hands if I didn't trust myself to do it. That way any hands the coach questions I can explain why I played them the way I did and expect the coach to understand why more then AI would. If I can't explain why I played I hand the way I did to a human coach then that would be 110% a leak. With AI you could never be 110% sure.

Cheers!!!
 
  • Like
Reactions: nilsharker
I think AI could probably do quite well if it actually asked you to explain your actions in a hand and analyzed what you told it, but that is assuming you can actually communicate the reasoning behind your hands in a coherent manner and compatible with the AI.
 
Yes, and the stats side of it is not even new. Trackers have built leak reports out of your database for years: fold to c-bet by position, big blind defence against each open size, how often you get to showdown and win there, each set against a range that winning players sit in. What AI adds is the plain-language part, turning a table of numbers into "you overfold the river to big bets". The catch is sample size. Preflop numbers settle after a few thousand hands, but a river spot against one bet size comes up so rarely that a tool can flag a leak that is really just variance. So I'd trust it as a pointer to where to look, then go through the actual hands in that spot before changing anything.
 
G0930 said:
I prefer a human being as a pokercoach.
But yes if you feed AI with a big enough hands sample it surely can detect flaws
That's a fair point! A good human coach can understand the reasoning behind your decisions in a way that AI might struggle with.

I think AI could be especially useful between coaching sessions, though. Imagine having something that continuously identifies recurring mistakes, so when you sit down with your coach, you already know which areas need attention.

Would you see AI as a useful complement to a human coach, rather than a replacement?
 
eberetta1 said:
It would be interesting to see if AI could win a poker tournament with the cards it is dealt, much like a machine beating a chessmaster at chess.
That would definitely be interesting to watch! 😄

I think the biggest challenge would be seeing how well AI adapts to different opponents throughout a tournament, rather than just making theoretically optimal decisions.

Especially when stack sizes, tournament pressure and player tendencies start changing!
 
patchworkfox said:
Yes, and the stats side of it is not even new. Trackers have built leak reports out of your database for years: fold to c-bet by position, big blind defence against each open size, how often you get to showdown and win there, each set against a range that winning players sit in. What AI adds is the plain-language part, turning a table of numbers into "you overfold the river to big bets". The catch is sample size. Preflop numbers settle after a few thousand hands, but a river spot against one bet size comes up so rarely that a tool can flag a leak that is really just variance. So I'd trust it as a pointer to where to look, then go through the actual hands in that spot before changing anything.
Really good point, especially about sample size! 👏

I completely agree that traditional poker trackers have been identifying statistical leaks for years. The interesting part is what AI could add beyond simply translating those numbers into plain English.

Imagine a system that identifies potential leaks, explains the GTO reasoning behind them, and then lets you practise those exact situations to improve.

That's actually something we're working on at OptimusPoker, combining solver-based training with AI-powered analysis and personalised leak detection.

And you're absolutely right about variance. A small sample shouldn't automatically be treated as a confirmed leak.

Would you find it more useful if AI automatically generated training exercises based on your potential leaks, rather than just giving you a report?
 
hardongear said:
No I wouldn't trust AI. AI only knows perfect poker math and GTO. The list of what AI doesn't know far out weights the list of what it knows.

AI doesn't know player pool tendances, doesn't know exploitive play, doesn't have instincts, doesn't have player reads and doesn't know when or how to bluff. Without having/knowing all them things AI can't tell the difference between a leak, an exploitive play, a bluff or instincts.

Now if one hand a million+ hands to feed AI and knew how to train AI, and knew how to ask AI the right questions because AI can be straight up be wrong and be mislead depending on the way/wording of ones questions seen it many times.......I'd say maybe.....

Even then I'd still pay a coach to review my hands if I didn't trust myself to do it. That way any hands the coach questions I can explain why I played them the way I did and expect the coach to understand why more then AI would. If I can't explain why I played I hand the way I did to a human coach then that would be 110% a leak. With AI you could never be 110% sure.

Cheers!!!
You actually make some really good points here, especially about distinguishing between a genuine leak and an intentional exploitative adjustment.

I agree that blindly trusting AI would be a mistake. And there's definitely a difference between a generic chatbot giving poker advice and a system using actual solver calculations to evaluate decisions.

Even then, a deviation from GTO isn't automatically a bad play. If you're exploiting a specific opponent, that deviation might be completely justified.

I think the interesting challenge is whether AI can learn to distinguish between repeated mistakes and intentional adjustments, especially when you give it context about why you made a decision.

That's also why I think AI should explain its reasoning rather than simply label decisions as mistakes.

Would your opinion change if an AI tool showed you the exact solver calculations behind its conclusions and allowed you to explain your reasoning for each decision?
 
  • Like
Reactions: hardongear
optimus.poker said:
You actually make some really good points here, especially about distinguishing between a genuine leak and an intentional exploitative adjustment.

I agree that blindly trusting AI would be a mistake. And there's definitely a difference between a generic chatbot giving poker advice and a system using actual solver calculations to evaluate decisions.

Even then, a deviation from GTO isn't automatically a bad play. If you're exploiting a specific opponent, that deviation might be completely justified.

I think the interesting challenge is whether AI can learn to distinguish between repeated mistakes and intentional adjustments, especially when you give it context about why you made a decision.

That's also why I think AI should explain its reasoning rather than simply label decisions as mistakes.

Would your opinion change if an AI tool showed you the exact solver calculations behind its conclusions and allowed you to explain your reasoning for each decision?
Thanks I tried my best was hoping I was making some sense.....my wording and phrasing isn't always too great.

That's an interesting question and I honestly don't know.....as of right now currently AI can be fooled, give questionable and straight up wrong answers in certain situations just based off how one asks a question or their wording. Where as human coach who truly knows the ins and out of poker doesn't have that problem.

Fwiw there's also not many poker coaches I would honestly pay money for their services either. As personally I generally feel the majority of those who now coach fall under the those that can't do anymore coach if you get my drift. And why would anyone pay for a coach that can't win anymore? If I was to pay anyone for personal coaching it'd honestly Phil Galfond just because of the type person he is on the felt and off. He also don't take students just because you have the money to pay him.......he has certain level of talent he needs and wants to see in his students. He doesn't see what he likes he doesn't take you as a student. He's simply not in for just the money and he still can win at the highest level.

Cheers!!!

Cheers!!!
 
optimus.poker said:
That's a fair point! A good human coach can understand the reasoning behind your decisions in a way that AI might struggle with.

I think AI could be especially useful between coaching sessions, though. Imagine having something that continuously identifies recurring mistakes, so when you sit down with your coach, you already know which areas need attention.
AI is especially useful when you give it precise orders and you get your information you want, within seconds .

Sparing time is the convenciency but if we rely on it , it will numb our brains for sure .

Theres a saying :"The brain is a muscle, train it or lose it"
optimus.poker said:
Would you see AI as a useful complement to a human coach, rather than a replacement?
Definetly complement rather than replacement.
Though in many aspects of life outside of poker, some replacements are inevitable..
But that's another story
Back to poker.
If you ask it for example to analyze your hands sample of let's say 100K hands how you performed based on GTO , it will give you precise results.
Which you can ofc get from a coach or paid App but it will take much longer.
Gemini e.g is for free, coach/gto app isn't
 
AI can work out a lot of things for you.
It can make really good analysis of your player pool if you have a large enough hands database.
On this MDA it is able to work out some exploits that may help your winrate.
 
Definitely, especially when it comes to frequency leaks rather than obvious single-hand mistakes. Most of us notice when we make a bad river call, but it’s almost impossible to feel on your own that you’re under-defending the big blind by 8% against button opens or c-betting dry boards too often. Feeding a decent sample of hand histories into solvers or leak-trackers highlights those blind spots immediately—the real challenge is actually translating those adjustments into live reads without turning into a robot.
 
been running LLMs over my own hands to watch how they explain their decisions, and the surprise: the model makes a fine fold and gives a confident reason that's plain wrong, or plays a spot right for the wrong logic. a leak-finder that reads your reasoning will miss the leaks where your logic sounds clean but the line is -EV. you still want the solver baseline underneath, with the model narrating the diff.
 
"No, I won't let AI ruin my playing style. Even if I lose, at least it's from decisions I made myself, not an AI.


However, I agree with having AI act as my analyst—that will improve my skills."
 
Back
Top