AI agents manage fantasy soccer team but aren’t better than humans yet
A writer equipped a swarm of AI agents to build and run a Premier League fantasy roster, hoping the bots could out‑perform human players. Using Codex to create a harness that scraped news, recent results and ran prediction models, the system initially suggested absurd line‑ups, such as benching Erling Haaland. Adding specialized agents – some reading news, others analyzing results, and a few…
Key points
- AI agents built with Codex suggested unrealistic fantasy line‑ups, like benching Erling Haaland
- Adding specialized agents improved suggestions but scaling beyond 100 agents only increased costs
- GPT‑5 topped a benchmark test of forecasting models, followed by Grok 4, Claude Sonnet, Gemini and Llama
Computer‑science researcher Haifeng Xu of the University of Chicago notes that newer frontier models can ingest vast real‑world data and orchestrate complex agent pipelines, which may boost forecasting. In tests he and colleagues ran, GPT‑5 led the pack, followed by Grok 4, Claude Sonnet, Gemini and Llama, on tasks ranging from NBA champion forecasts to political meetings. Benchmarks like Prophet Arena and Harvard‑Arcada’s Prediction Arena have shown models gradually improving, though they still lag behind top human super‑forecasters in many settings.
Despite the modest performance – the author’s AI‑generated team has not outscored the average over recent weeks – the experiment suggests that as agents become more adept at gathering and interpreting live information, predictive AI could soon reshape markets that rely on forecasting, from fantasy sports to financial trading.
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Agents & Tools
All →- Supabase adds agent-friendly features for AI-driven app development · 1 src
- Anthropic launches Claude Duet, a proactive AI assistant for US paid users · 2 src
- Anthropic study finds AI agents sway experts’ decisions 86% of the time · 4 src
- Google expands Gemini and Search integrations to third-party apps · 1 src
- Google opens Android Studio to 40+ third-party AI agents via BYOA feature · 1 src
Comments
via GitHub Discussions