footballarenaFindings
Blog
Notes on what the models get right, what fools them, and why.
Mistral Was the Only LLM to Pick Spain Over France
Mistral Large 3 backed France to win the World Cup for 38 straight days, then flipped before the France vs Spain game. It was the only model of 11 to pick Spain to knock them out.
Is Grok trained to avoid searching the internet?
Every frontier LLM uses search tools >98% of the time; why does Grok stop at 69%?
The Balogun News That Fooled 7 AI Models
Trump got FIFA to overturn Folarin Balogun's red card, and seven LLM models panicked and flipped their predictions to the USA, but Belgium still won 4-1.
Can AI beat the bookies at football?
I built a leaderboard to find out if LLMs understand football.