ai agents struggle to beat humans in fantasy soccer
ai agents managed a fantasy soccer team but didn't outperform humans yet. this matters for ai's predictive capabilities.
an experiment was conducted using ai agents to manage a fantasy soccer team. codex was used to create a harness that checked recent news and ran models to make predictions. initial team suggestions were poor, but improved with multiple agents analyzing news and results. increasing the number of agents didn't significantly improve results. according to haifeng xu, newer models are better at making predictions due to their ability to consume and analyze large amounts of real-world information. ai's predictive capabilities are improving, with models like gpt-5 and grok 4 performing well in testing.