Hi Sam! Aside from being a POTD Premium subscriber, I’m also a machine learning engineer. I’ve experimented with many ways to use LLMs to support my poker study, and quite a few of them haven’t worked.
For example, I tried having an LLM consume all my study notes and build a personal poker knowledge base. However, giving it more poker-related material never fully solved the hallucination problem. I think the underlying issue is that much of the available poker content to train LLM is inherently noisy. LLMs have also learned from poker discussions on online forums, twitier, etc, where the analysis is not always accurate or precise.
However, I’ve found one application that works surprisingly well, and I’d love to share it with you: using the LLM as a “poker student.” Its primary role is to read what I’ve written and ask questions as a student, rather than generate poker strategy itself, where hallucinations are much harder to avoid.
Here is an example of a study session I had with my LLM poker student. I assigned it the role of a skeptical peer student:
During the session, the LLM challenged an example I had used in another referenced principle—one involving betting large with a linear range. Working through its questions helped me understand the underlying concepts more clearly. I eventually crossed out the incorrect example and added a better one:
I remember in school we had an assignment to define words, but they were all words that were common in an eight grade vocabulary. The idea was that by defining or explaining something for someone who did not understand what you were trying to do, you'd end up bolstering your own understanding. This seems like conceptually the same thing. You have a naive student ask you "why?" and if you can't explain it well, it means you don't fullly understand why yourself.
Exactly. An LLM is especially useful because you can specify precisely what prior knowledge it should have for the task (ofc, it can’t always play the role of a true expert in poker currently).
This is also why I’ve been writing the “principles” in the example I shared. Writing is one of the best ways to learn: generating the words yourself strengthens encoding, connecting ideas to existing knowledge creates more retrieval paths (cross-reference related notes), and articulating them for a listener may be the highest-leverage version. The struggle to explain clearly is itself part of the learning.
You've been doinng all that by writting POTD already!
Hi Sam! Aside from being a POTD Premium subscriber, I’m also a machine learning engineer. I’ve experimented with many ways to use LLMs to support my poker study, and quite a few of them haven’t worked.
For example, I tried having an LLM consume all my study notes and build a personal poker knowledge base. However, giving it more poker-related material never fully solved the hallucination problem. I think the underlying issue is that much of the available poker content to train LLM is inherently noisy. LLMs have also learned from poker discussions on online forums, twitier, etc, where the analysis is not always accurate or precise.
However, I’ve found one application that works surprisingly well, and I’d love to share it with you: using the LLM as a “poker student.” Its primary role is to read what I’ve written and ask questions as a student, rather than generate poker strategy itself, where hallucinations are much harder to avoid.
Here is an example of a study session I had with my LLM poker student. I assigned it the role of a skeptical peer student:
https://github.com/sherryxiao1988/poker-student/blob/main/CLAUDE.md
I then asked it to read a poker principle I had written:
https://github.com/sherryxiao1988/poker-student/blob/main/study-sessions/2026-05-11%20-%20Bet%20size%20signals%20range%20polarity%20in%20practice/Bet%20size%20signals%20range%20polarity%20in%20practice.md
Here is the complete study session:
https://github.com/sherryxiao1988/poker-student/blob/main/study-sessions/2026-05-11%20-%20Bet%20size%20signals%20range%20polarity%20in%20practice/2026-05-11%20-%20Bet%20size%20signals%20range%20polarity%20in%20practice.md
During the session, the LLM challenged an example I had used in another referenced principle—one involving betting large with a linear range. Working through its questions helped me understand the underlying concepts more clearly. I eventually crossed out the incorrect example and added a better one:
https://github.com/sherryxiao1988/poker-student/blob/main/study-sessions/2026-05-11%20-%20Bet%20size%20signals%20range%20polarity%20in%20practice/Range%20polarity%20and%20sizing%20are%20distinct%20equilibrium%20dimensions.md
You might be able to do something similar with POTD — have an LLM read the content and ask thoughtful, skeptical questions about it. :)
I remember in school we had an assignment to define words, but they were all words that were common in an eight grade vocabulary. The idea was that by defining or explaining something for someone who did not understand what you were trying to do, you'd end up bolstering your own understanding. This seems like conceptually the same thing. You have a naive student ask you "why?" and if you can't explain it well, it means you don't fullly understand why yourself.
Exactly. An LLM is especially useful because you can specify precisely what prior knowledge it should have for the task (ofc, it can’t always play the role of a true expert in poker currently).
This is also why I’ve been writing the “principles” in the example I shared. Writing is one of the best ways to learn: generating the words yourself strengthens encoding, connecting ideas to existing knowledge creates more retrieval paths (cross-reference related notes), and articulating them for a listener may be the highest-leverage version. The struggle to explain clearly is itself part of the learning.
You've been doinng all that by writting POTD already!