ChatGPT and poker strategy
Do any of you find ChatGPT to be helpful for discussing poker strategy and hands?
I recently signed up for the $20/month version, and have been discussing poker with it.
It seems to be able to discuss strategic concepts pretty well (as far as my own knowledge can assess it).
I run hands by it, and it seems to me to do ok with it's final recommendations for what action to take on a street. It's reasoning seems to make sense.
But then sometimes, actually too often, it mis-states key details when discussing the hand. For example, deep into a conversation about a hand, it will say something like, "the villain could hold KQ for the nut straight..." when the board we are discussing is A T 9 5 3. However, in this situation, it still seems to give a decent recommendation on what I should do on the river with my hand.
I sometimes chat with it during sessions to review hands or concepts, or to stay focused on my strategy.
Here are ChatGPT's pre-session coaching tips for me:
Pre-Session Mental Game Reminders
Stay patient.
Don’t force the action—let edges come to you.
One pair is one pair.
Even QQ or KK isn’t an automatic stack-off. Play the board, not just your hand.
Avoid ego plays.
Don’t try to "prove" something with a hero call or bluff. Win chips, not points.
You don’t need to win the session in one hand.
Avoid forcing big pots early—build your image and rhythm first.
Take pride in disciplined folds.
You’ve been winning despite mistakes. Fewer punts = more profit.
📈 Technical Strategy Reminders
Don’t bluff-raise when calling with a bluff-catcher is better.
Only raise when you know what you’re repping and what folds.
Don’t shove rivers when a smaller value bet will get paid more often.
Get the call, not the fold.
Use blockers—but don’t rely on them too much.
They’re a tiebreaker, not a green light.
Use population reads.
Turn check-raises = strong
River overbets = value
Cold-calls = small pairs or suited Broadways
🎯 Execution Prompts to Ask Before Big Decisions
What’s my opponent’s range here?
What worse hands call me if I bet?
What better hands fold if I bluff?
What do I gain by betting instead of checking?
Would I stake someone else to make this play?
📝 Session Management Tips
Take mental notes every orbit or two:
Who’s sticky? Who’s tight?
How do they react to aggression?
What do they think you’re capable of?
Label your own table image.
Ask yourself every 30–45 minutes:
“How would I describe myself if I were watching this game?”
Watch for time pressure moments.
Take 3 seconds before any river bet or call. Ask:
“Is this a bet… or a flex?”
7 Replies
Do any of you find ChatGPT to be helpful for discussing poker strategy and hands?
It seems to be able to discuss strategic concepts pretty well (as far as my own knowledge can assess it).
But then sometimes, actually too often, it mis-states key details when discussing the hand.
The main thing you need to know about LLM "AI" is... that it's all a scam.
It doesn't think, it doesn't know.
At best it's taking every post on here and searching for the "best matches" given the words you input and spewing out some random collection of the words used to answer those posts, more likely you'll get the "10:10 Phenomenon" generic answer reworded from "you posted, so you lost, so fold" all the time and it'll be correct because people don't fold enough at low stakes.
I understand that it's a bit confusing because GTOw doesn't "know" what it's doing, in a human sense. But it does know the rules of poker and has "played" so many hands it has a known result for what it would do. Generative AI knows nothing.
I would be careful with it. There have been several cases where lawyers got in trouble using AI to write legal arguments, and it made up cases.
LLMs aren't a scam, they just work off off next token prediction
meaning based on the words you entered before, they predict what the next word is supposed to be, and the next word after that
it works astonishingly well for a wide variety of use cases
but for poker probably not that well
gives me an idea for how they could be improved, but poker studying is such a small market it's not really worth pursuing
Request to mods to pin this thread and all my brilliant insights therein until the rate of questions about GenAI's unsefulness in poker training slow down: forumserver.twoplustwo.com/170/live-no-limit-holdem-cash/asked-deepseek-poker-strat-heres-what-i-got-1844931/
Some highlights from yours truly:
To the extent that LLMs are useful for producing poker strategy*, you will have to do a LITTLE more work then sending a 200 character prompt and copy and pasting whatever it spits out. It takes some prompt engineering, proofreading, follow up prompts, etc.
*Which, at least for now, is limited to things like building out blog content or formatting a hand history.
And
AI is an extremely broad term with a lot of specialized applications for smart people who are very thoughtful about it. PokerSnowie is built entirely on AI, and obviously that was quite revolutionary for poker when it came out. GTOW has AI-enabled functionality that (AFAIK) is extremely useful.
"What AI is decidedly not useful for is typing lazy prompts into a mid-tier chatbot and copy and pasting the output to share with people who have specialized knowledge in the field as if it's going to provide us anything other than punchlines.
And
[AI can do all sorts of things, but {DeepSeek / Gemini / ChatGPT / Copilot / Claude /etc} are] a specific type of AI, namely a large language model chatbot, which scrapes the publicly available web, aggregates it and gives a language-based response.
I am sure the AI functionality in GTOW is programmed to scrape and process the data of things like hand charts, as well as the frequency, EV, and equity views, etc and aggregate THAT, likely producing outputs that use GTOW's dashboards, views, reports, etc. This is extremely useful because while we can "just use a solver ourselves", most of what using a solver means is learning how to process a great deal of information and translate it into some framework that's digestible and implementable to our human brains (which is why different views and dashboards exist in the first place.)
To the extent that purely language-based interaction using publicly available information is useful, you do not have to just get a shotgun spray of randomly available information on the entirety of the web and can at least get it to analyze your HH as if it were UpSwing poker or submersible or what have you.
Just a few weeks ago I asked Gemini about a term in some poker training materials I was unfamiliar with, which you'd think would be right up its alley because essentially a troubleshooting issue and the answer should be somewhere on the publicly available web. Instead, it very confidently dug its heels in on a (completely incorrect) guess. It even got defensive when I asked if they were basing this on anything other than context clues.
I later realized it was a translation issue, and Gemini was at least helpful in helping me figure out what the word in the original language meant in that country's poker slang.
So not completely useless, but even for this very specific use case it required me to do at least as much work as the machine to get anything not misleading, much less correct.
It will probably get better at this sort of thing over time, but LLM-based GenAI just flatly isn't the right tool for most poker study for anyone with intermediate+ knowledge.
I don't trust GPT & AI in general. Especially after I uploaded a photo of my pitbull & asked it what breed it thought it was & it replied it was a Yorkshire Terrier, lol.
