The companies Google DeepMind and Kaggle brought together eight of the leading Large Language Models (LLMs), more commonly known as artificial intelligences, and had them compete in the Kaggle Game Arena in three different disciplines: poker, chess, and Werewolf.
Liv Boeree
OpenAI in Command
The game began this Monday, February 2nd, with the first round. The LLMs started in the quarterfinals and had to overcome a rival in each stage. Day 2 was played on February 3rd, and the final is being played today, February 4th, between o3 and GPT 5.2, two different products from the same company: OpenAI.
This is not the first time a poker game between LLMs has been organized. In October 2025, Max Pavlov
The Voice of the Experts
The first to generate content on the matter was Doug Polk, who conducted an exhaustive analysis of the hands played by the AIs. Polk not only analyzes the hands from a human perspective but also attempts to explain the decision-making process carried out by the LLMs. In many cases, the AIs failed to perform logical reasoning.
According to what he himself commented on his X account, “LLMs do not understand suits or suited draws. My favorite hand was when GPT-5.2 thought it had a draw with 8 6 on a board of Q J 5 3.” It’s curious because it’s a mistake often found in humans who are just learning the game.
Read also: Samuel Vousden Crowned GGMillion$ Champion
On Day 2, Nick Schulman and Nakamura met in a stream to analyze the hands played by the AIs. Nick has considerable experience in the commentator’s booth and is good at using his game knowledge to analyze strategies.
It’s interesting to see Schulman trying to rationalize the decisions made by the LLMs, which, as we saw with Polk, often failed in their reasoning.
Finally, Liv Boeree took the task not only of analyzing the game itself but also of questioning the fact that AIs play. Her video is titled: “Why Did Google Make ChatGPT, Gemini, and Claude Play 900,000 Hands of Poker?”, posing a question from the outset.
“I can imagine that some people find it concerning, or at least worthy of concern, the fact that we are training LLMs on decision-making games like Werewolf and poker, as we might be incentivizing persuasive and manipulative behavior in LLMs,” stated Liv, who has long been denouncing the dangers of technology and modern life.
“I think it’s a valid concern. What do you think? Do you agree with this, or do you not think it’s the right way to frame this issue?” concluded the player, inviting her followers to reflect on it.
Read also: “MatasSembolas” Won The Venom Mystery Bounty
