Grok 4 Picks the Dodgers to Win the World Series—But Other AIs Aren’t So Sure
Elon Musk’s xAI had its big Grok 4 showcase this week, and one of the flashier demos involved asking the AI to predict the 2025 World Series winner. After chewing through Polymarket data for about four and a half minutes—using what xAI calls its “Heavy” reasoning mode—Grok spat out its answer: the Los Angeles Dodgers, with a 21.6% chance. Not exactly a lock, but better odds than anyone else. Still, Grok hinted the Dodgers might be overvalued.
The prediction isn’t out of left field. Over at ESPN BET, the Dodgers are sitting at +225 as the season rolls toward the All-Star break. Meanwhile, the Detroit Tigers, leading the AL Central with a 59-35 record, are getting buzz as a dark horse at +750. Some folks on X are already joking that Grok could be an “infinite money glitch.”
But here’s the thing—when we asked other AI models, they didn’t all agree.
ChatGPT, DeepSeek, and the Case for Detroit
ChatGPT’s latest model gave the Dodgers a slightly higher chance (26%) but also called them overpriced. It liked the Tigers better as a value bet, putting their win probability at 16% against market odds that implied just 12.5%. The reasoning? Detroit’s pitching, especially Tarik Skubal, has been lights-out this year.
DeepSeek was closer to Grok, favoring the Dodgers at 23%, but even it admitted the Phillies might be the smarter wager based on risk-reward.
Now, here’s where it gets interesting. We didn’t shell out for Grok 4’s $300/month “Heavy” tier—because, well, budgets exist—so we tried the cheaper $30 version instead. Shockingly, it gave the Tigers a tiny edge over the Dodgers, with less than a percentage point separating them.
All three models zeroed in on similar factors: Detroit’s pitching depth, the Dodgers’ injury woes, and the tendency for betting markets to overhype recent champs.
You Don’t Need a $300 AI to Get Useful Predictions
Grok 4’s “Heavy” mode is impressive, but you can get decent results without it. The key? How you ask. After playing around, we found three things that help:
1. **Role-play matters.** Tell the AI to act like an expert prediction analyst. Something like, *”You’re a Bayesian forecasting specialist with deep knowledge of sports betting markets”* works way better than a vague question.
2. **Methodology beats guesswork.** Ask the model to pull odds from multiple sources, compare them to stats, and flag discrepancies. Models do better with clear steps—what prompt engineers call “Chain-of-Thought.”
3. **Anchor predictions in real data.** Mentioning Baseball-Reference or FanGraphs keeps the AI from just making stuff up.
For fun, we built a custom GPT that mimics Grok 4’s approach—it scrapes odds, analyzes teams, and looks for value. The code’s up on GitHub if you want to tinker.
Just remember: This isn’t financial advice. If you win, great. If you lose… well, maybe don’t bet the house on an AI’s hunch. But hey, if it works out, beers are always welcome.
