Ivan vs the Machine: What Happened When I Put an AI Model Against a Sports Journalist
I built an AI model to predict every World Cup 2026 match against a sports journalist's gut calls — Monte Carlo, GPU on Solana, and a Hedge algorithm.
Monte Carlo simulation pits AI model against sports journalist for entire World Cup
For the 2026 FIFA World Cup, a Brazilian YouTube show and platform created a segment called Ivan vs the Machine. Every round, sports journalist Ivan Santos gave his prediction for each match, while a model built by the reporter predicted the same matches using only data. Over 38 matches, the results showed the AI model was correct 73.7% of the time, compared to the journalist's 50.0%.
The reporter had to build two architectures to achieve this feat: the first version used neural embeddings on a GPU but failed due to diverging training loss. The second version dropped the GPU and instead used a weighting algorithm based on two human-based signals: current squad value and in-tournament form. These were combined into a single strength score, which then fed into a Poisson expected-goals calculation.
To predict the match outcomes, the model drew 50,000 samples from two Poisson distributions and derived win/draw/loss probabilities. This process took under a second on an ordinary CPU, making it feasible to run the simulation for the entire World Cup. The weighting algorithm then treated Ivan's picks and the AI model's predictions as competing experts, with the model adjusting the weights of correct and incorrect predictions after each match.
Written by urgent.news from HackerNoon's reporting — not their text. Machine-written — it may contain errors, so check the original before relying on it.
