Metaculus has wrapped up the Fall 2025 FutureEval Bot Tournament. This tournament (formerly known as AI Benchmarking) is part of the FutureEval project, measuring the ability of AI agents to predict future outcomes in Science, Technology, Health, Geopolitics, AI itself, and more. Beyond tournaments on Metaculus, forecasting is a key skill in many real-world tasks, enabling planning, risk assessment, and decision-making.
As we’ve done after previous rounds, we surveyed participants about how they built their bots. 39 developers responded (29 prize winners and 10 non-winners), and we merged their answers with the final leaderboard and the per-bot behavior logs.
The takeaway: when everyone is using frontier models, scaffolding choices make a much larger difference.
FutureEval Forecasting Bot-Maker Survey: What Winners Did Differently
Link post
Metaculus has wrapped up the Fall 2025 FutureEval Bot Tournament. This tournament (formerly known as AI Benchmarking) is part of the FutureEval project, measuring the ability of AI agents to predict future outcomes in Science, Technology, Health, Geopolitics, AI itself, and more. Beyond tournaments on Metaculus, forecasting is a key skill in many real-world tasks, enabling planning, risk assessment, and decision-making.
As we’ve done after previous rounds, we surveyed participants about how they built their bots. 39 developers responded (29 prize winners and 10 non-winners), and we merged their answers with the final leaderboard and the per-bot behavior logs.
The takeaway: when everyone is using frontier models, scaffolding choices make a much larger difference.