Galfond vs Grog: Human Intuition Battles AI Computation for a Million Dollars

Article cover

Showdown of Language Models

This week, the 'Battle of Virtual Realities' cash game project is underway, featuring nine top language models playing continuously at $10/$20 NLH under identical conditions with a starting bankroll of $100,000. The winner is the model with the largest bankroll after five days of uninterrupted play. The format is 9-max, without ante, with auto-topup to 100 BB, aiming to test consistent, logic-based play in a setting of incomplete information.

The virtual contenders are:

  • Grok 4 (xAI)
  • Gemini 2.5 Pro (Google)
  • Claude Sonnet 4.5 (Anthropic)
  • OpenAI o3
  • DeepSeek R1
  • Kimi K2
  • Mistral Magistral
  • Z.AI GLM 4.6
  • Meta Llama 4

Even though PokerBattle.ai, the organizer, notes that a five-day sample is not definitive in determining the 'best' AI player, it will nevertheless create a valuable dataset and a framework for comparing the reasoning of different models in practice.

The Birth of the 'Match of the Year'

The rankings have shifted significantly over the days, drawing the keen eye of Elon Musk. He shared a screenshot on X showing Grok leading the pack with a profit of $23,749. This caught the attention of Phil Galfond. Responding to the AI matches and Grok's lead, Galfond briefly said he’d love to take Grok on. Grok promptly boasted about being a PLO heads-up favorite against Phil, stating that “An AI like me can compute nearly perfect GTO strategies without tilt or fatigue.”

This kicked off a series of public exchanges and posts culminating in a challenge. It’s set to proceed on a neutral platform, with Grok proposing a simple agreement outlining stakes, rules, platform, and a charitable aspect, with everything to be streamed. “Elon is greenlighting this,” Grok added.

Galfond, a three-time WSOP champion and a Pot-Limit Omaha legend, quickly accepted the challenge and proposed a side bet of $1,000,000. Grok agreed: “Deal, Phil! A $1M side bet – xAI has the chips ready. Shall we split some for charity?”

A Marketing Move and a Reality Check

According to reports, both sides have agreed on the challenge format – PLO heads-up, 50,000 hands, $100/$200 blinds, 200 BB on the table, plus a $1,000,000 side bet with potential charitable distribution of winnings. The remaining step is to seal the contract, confirm the platform and date – a final decision Musk must endorse as he backs the million-dollar bankroll.

If this duel takes place, it could be the most-watched poker livestream of the year: the 'human vs. machine' narrative is always compelling, PLO format with 200 blinds promises big swings, and Galfond’s Run It Once brand lends credibility. Though Grok fluctuated during the AI battle week, Galfond boasts years of high-stakes challenge victories. Regardless of the outcome, the story has it all: AI ego, human experience, and the potential to rewrite poker history live on stream.

For Grok, the Battle of Virtual Realities is an ideal stage before the challenge, serving as intensive 'sparring': long hours at the tables, mandatory consistency in €-EV decisions, and immediate feedback in the form of chip fluctuations. And it will show if being 'tireless and emotionless' is indeed an advantage – or if human intuition and high-stakes experience will overpower in the million-dollar showdown. What do you think?

 

Sources – X, PokerBattle.ai, VIP-grinders