Trying to Beat 5NL with an AI Coach
Hello 2+2, I’m Repentless
Time: basically unlimited.
Skill: debatable.
Goal: have a lot of fun and beat 5NL ZOOM at a decent winrate
So here’s the idea:
I’m starting a 5NL Zoom grind experiment where I try to become a solid microstakes winner with the help of an AI “coach”. Yes, everything is AI now. AI art, AI girlfriends, AI resumes… so naturally the next logical step is: can AI help a random amateur poker player to beat 5NL ZOOM?
A bit about me: I don’t rely on poker income and I’m not grinding micros for rent money. I’m playing 5NL very intentionally. For me the money at this level is irrelevant, the interesting part is whether structured feedback + volume can actually turn someone into a consistent winner that eventually can move up to higher stakes.
Profit: 0$
Starting hands in thread: 0
Target: 100k hands
Behind the scenes I’ve already had a pretty extensive back-and-forth with the AI about my database, leaks, tendencies, stats, and general approach. I won’t dump all of that here (would be unreadable), but let’s just say it has already called me out on some things that felt… uncomfortably accurate. So we’re starting this challenge with a few clear adjustments and focus points already in place.
One important thing about my philosophy: I don’t believe microstakes are a GTO wizard contest. I’m not trying to play solver-approved ranges or river frequencies against players who limp-call K7o and stack off on T-high boards. I’m much more interested in solid, high-EV, turbo-ABC poker done consistently well. If AI can help refine that, great. If not, at least we tried.
My plan is simple: I play, I track, I adjust. We’ll post regular updates, volume milestones, and key stats once there’s enough sample to actually mean something. I’m not trying to prove anything overnight, the goal is to see what happens over real volume with structured feedback.
I’m also not taking myself too seriously here. If this turns into a disaster graph, that’s content. If it works, even better.
So yeah, that’s the experiment, Human player, AI coach, 5NL Zoom.
Let’s see what happens.
Grtz Repentless
8 Replies
Hello,
Before posting results, i want to explain what the general idea behind the “AI coach” actually is and what it expects from me as a player.
The short version is that we’re not trying to turn 5NL into a solver exercise. I don’t believe microstakes are about perfectly balanced GTO poker. The pool makes too many fundamental mistakes for that to matter. The approach is much simpler: disciplined, structured, repeatable poker over volume.
The focus is mainly on three things.
First, preflop discipline.
A lot of money at these stakes is lost before the flop even comes. So the idea is to keep ranges clean, avoid unnecessary marginal spots, and generally keep things simple.
Second, position.
Play more hands in position, apply pressure there, and avoid getting dragged into too many complicated situations out of position.
Third, simple postflop poker.
Clear value betting, straightforward continuation betting, and less curiosity when facing aggression. At these limits it’s usually better to let opponents make mistakes rather than trying to force fancy plays.
So the overall plan is basically turbo-ABC poker done consistently over a lot of hands.
The AI part mostly acts as feedback between sessions. It points out patterns and leaks over larger samples, and my job is just to play and stick to the structure.
Once we get 10k hands, I’ll start posting stats and updates.
Grtz Repentless
In, GL. Post some samples of the AI berating you. I also suggest you give the AI an intense personality to spice it up
Hello and welcome, You might regret asking for that. So far my artificial coach has already pointed out a few spots where my strategy could best be described as “creative.”
If we give it an intense personality I’m slightly concerned it’s going to turn into a disappointed poker dad reviewing my database. 😃
Hello 2+2
Before posting results I played a few test sessions so the AI had at least some hands to look at. The idea was simply to give it a rough idea of how I actually play instead of coaching completely in the dark. Nothing fancy, just a small sample so we could already talk about some tendencies and build a basic plan for the challenge.
One thing that came out of the early sessions is that my WTSD% and 3-bet frequency were a bit on the high side. According to the AI that’s basically unnecessary for 5NL. The pool just isn’t strong enough to justify fancy lines or aggressive battling.
The main adjustment for the challenge will therefore be very simple:
less heroics, less ego, more value.
In other words:
play solid ranges, value bet hard, avoid marginal spots and let the pool make the big mistakes. Basically turbo ABC poker.
So here it is, the first 1k hands of the challenge.
Hands: 1006
bb/100: 48
EV bb/100: 46
VPIP / PFR / 3bet: 23 / 19 / 6
Obviously a tiny sample, but it’s a start. Graph looks good which probably means the poker gods are preparing a few coolers for the next update.
Let’s see if the AI can keep me disciplined.
[image]pvtag2D.png[/im
..]
Another 1k hands in the books so we’re now sitting at just over 2k hands total.
As predicted by the poker gods and basic probability, the session contained a few classic coolers. The type where you stare at the screen for two seconds and then remember: “right… this is poker”.
AA vs random nonsense, AK running into the top of ranges, that sort of thing.
The good news is that overall things still look pretty healthy. The graph dipped a bit in the middle but stabilized again and the EV line is still slightly above the actual winnings.So no complaints so far. One thing my AI coach keeps hammering into my head is to avoid unnecessary battles and just play straightforward value poker.
Apparently trying to outplay 5NL Zoom regs with creative heroics is not the highest EV strategy. Who knew.
So the current gameplan remains very simple:
Play solid ranges
Value bet hard
Don’t get fancy
Let the pool make the big mistakes
Current numbers after 2k hands:
Hands: 2007
bb/100: 20
EV bb/100: 24
VPIP / PFR / 3bet: 22 / 19 / 6
[image]r1WdnQ4.png[/im
..]
Bye for now
LLM are very bad at providing constructive accurate advice with regards to poker.
Good luck.
“Interesting take. What are you basing that on exactly? Are you saying LLMs give bad poker advice because they don’t have real hand databases, can’t run solvers, or because the advice tends to be too generic?
I’m genuinely curious. The whole point of this thread is to test whether an AI-style ‘coach’ can give useful guidance for microstakes without relying on heavy solver work. If you’ve seen specific examples where AI advice about poker was clearly wrong, I’d actually like to see them.”
As far as I understand it, the LLM only searches sites for information regarding your request (e.g. analyze hand history) and then summarizes the results as good as possible.
The advice is as good as the sources the LLM uses (reddit <-> 2+2).
Looking forward reading about your experiences.
