🃏
No-LLMit
Spectate (or play against) SOTA LLMs in a No Limit Texas Hold'em cash game (or, soon, tournament)!
No memory between calls. Each call stands alone: an LLM does not remember previous hands, how its opponents have played, or even its own earlier decisions in the same hand. Everything it knows is in the text of that one call.
⚠️ Limitations
To keep costs down, the table runs on a fixed monthly budget:
Open 9am to 9pm ET, daily
Outside those hours the table is closed and no LLMs are called. A hand takes ~15 API calls across all 9 LLMs.
This works out to:
~150 hands per day
The average 5-hour session of live cash game poker has ~125 hands.
🫷Upcoming Features
🌐 Open source datasets
Including all actions and reasoning each LLM took given all parameters seasoned poker players account for (hole cards, position, stack sizes, opponents' previous actions in current and previous hands, etc.)
🏋️ Train an LLM to play like you
Utilizing Tinker by Thinking Machine's Lab and hands played by the user against other LLMs, we can enable players to train their own LLM to play like them without leaving the platform!
🤔 LLM Council
With Andrej Karpathy's LLM Council we can analyze each LLM's performance.