Grok 4 Predicts Dodgers for World Series Win—But Other AIs Aren’t So Sure

Among the demos Elon Musk showed off during Grok 4’s launch on July 9 was a banger asking the AI to predict which team will win Major League Baseball’s World Series later this year.

After 4.5 minutes of number-crunching that analyzed data from Polymarket, the Ethereum-based prediction markets platform, and using what xAI calls its “Heavy” reasoning capabilities, Grok 4 delivered its verdict: The Los Angeles Dodgers are the most likely team to win the 2025 World Series. Grok gave L.A. a 21.6% chance to win it all—higher than any other team, but still noted they might be overpriced.

Grok’s predictions are certainly in line with other major platforms, including ESPN BET, which shows the Dodgers sitting at +225 as the MLB season approaches the All-Star break. The Detroit Tigers (+750), who are running away with the AL Central, have emerged as a dark horse contender with baseball’s best record at 59-35.

Traders on X are giddy about the potential of having a personal Grokstradamus and calling the results an “infinite money glitch.”

But we wanted to know: Did the other major AI models agree with Grok?

Turns out, not entirely.

What other AIs think

ChatGPT’s o3 model gave the Dodgers a 26% chance while flagging them as overpriced. The model identified Detroit as offering the best value with a 16% win probability against market odds implying just 12.5%. Its reasoning centered on Tigers ace Tarik Skubal’s dominance and the team’s league-best pitching staff.

DeepSeek doubled down on Los Angeles with a 23% probability, but noted the Dodgers might be riding too much positive sentiment. Despite favoring LA to win, the model said it would rather bet on the Phillies because the risk-to-reward ratio was more compelling.

Since we’re poor and our paymasters were unlikely to approve Grok 4 Heavy’s $300 subscription for just one question, we asked the lighter Grok 4 version available via the $30 tier. Interestingly, it gave the Tigers a razor-thin edge over the Dodgers—less than one percentage point separated their odds.

All three models flagged similar factors: Detroit’s elite pitching rotation, the Dodgers’ injury concerns, and historical patterns suggesting the market overvalues defending champions.

It’s all in the prompt

While Grok 4’s “Heavy” reasoning is impressive, you don’t need a $300/month plan to get solid predictions. With smart prompting, even basic models can deliver sharp insights. We found that successful prompts need at least these three main elements:

First, role-play. Tell the model who it should be and how it should act. Try something like: “You are an expert Prediction Market Analyst with deep knowledge of Bayesian forecasting and risk management.”

Second, the methodology: Tell the model what you want and what steps it should follow in order to succeed. Ask the model to gather current betting odds from multiple sources, compare them against analytical projections, and identify value bets. Models perform better when they can compare market consensus against their own calculations.

This is what prompt engineers call Chain-of-Thought—if the model knows exactly what to do, it provides better results. Don’t know how to guide it? Ask the model separately for the steps needed to complete your task.

Third, point toward analytical resources. Mentioning Baseball-Reference simulations or FanGraphs projections helps ground predictions in established frameworks, rather than pure speculation.

For those interested in trying this themselves, we built a custom GPT that replicates what xAI demonstrated with Grok 4. It was just a fun experiment, but it gathers odds, analyzes team performance, and identifies potential betting value through natural conversation.

We also tossed our prediction market prompt on GitHub if you want to experiment with your own chatbot.

Use at your own risk, naturally. We’re not financial advisors, and neither are these AIs. If you lose, don’t blame us—but if it helps you win big, then we won’t say no to a beer.