RUC's EvoForesight Tops Real-Time Forecasting Leaderboard FutureX for Three Weeks
智东西 · wechat · 2026-10-23
Renmin University's IQuest lab introduced EvoForesight, a self-evolving framework for real-world event prediction that has topped the weekly FutureX leaderboard for three straight weeks. It closes the loop between three modules: Prometheus generates fresh, verifiable prediction tasks from 125 websites daily; the Branchus agent predicts via a evidence-gathering team and a judgment team with full reasoning traces; and Athena improves the system along three axes — RLVR into weights, distilling reusable skills, and evolving the environment to target the agent's weaknesses. Branchus scored highest on four different base models (Qwen3.8-Max, GPT-5.6-Sol, DeepSeek-V4-Pro, Claude-Opus-4.6), sweeping the top three overall. With just 88 real trajectories it improved markedly, and it produced near-correct public forecasts for the 2026 Nobel Prizes in Physics, Peace, and Literature, including lesser-known laureates. Paper and code are open-sourced.
More from AGI Musings
- AI economist ranks grew 10-30x, yet early insider says most newcomers' frames are useless or harmful — voooooogel · 2026-10-24
- Sequoia argues the next trillion-dollar company will sell services, not software — 2C_ornot2C · 2026-10-24
- Paper models LLM adoption as a 'cognitive virus' with population-level tipping points — 2C_ornot2C · 2026-10-24
- Microsoft AI CEO says agent swarm progress took the whole industry by surprise — anirbanbandyo · 2026-10-24
- The ChatGPT moment for robotics was…ChatGPT itself — eigenron · 2026-10-24
- "Is a virus conscious?" The one-liner puncturing AI risk debates — AlexTensor · 2026-10-24