AI Superforecasters Near Human Parity, Turn $35 Into $2 Million on Kalshi
Updated
Updated · astralcodexten.com · Jul 2
AI Superforecasters Near Human Parity, Turn $35 Into $2 Million on Kalshi
1 articles · Updated · astralcodexten.com · Jul 2
Summary
Metaculus results and startup claims suggest AI superforecasters are now roughly level with top humans, with one recent Cup placing a Preseen bot third behind two people.
Scaffolded systems appear to close much of the remaining gap: Metaculus said forecasting-focused setups are worth about 9 months of base-model progress, putting top AIs near pros’ scores of 31 versus 36.
Profit stories are fueling the excitement. One founder said his AI grew $35 to $2 million on Kalshi in seven months, while another claimed a market-neutral portfolio beat the stock market by 25%.
A live test showed why the tools are drawing attention: FutureSearch spent 5 minutes, used 3 subagents and 212 sources, and priced the chance of halving US respiratory infections by 2040 at 7%.
If current improvement rates hold, bots could move beyond human forecasters within a year, with finance likely to shift first and prediction markets becoming more AI-driven infrastructure.
When AI can turn pocket change into millions, who truly profits: the everyday user or a new AI-powered elite?
As AI forecasters dominate markets, what invisible 'epistemic drift' risks could trigger the next global crisis?
From Crowd-Level to Superforecaster: The Rise of AI in Large-Scale Forecasting (2024–2026)
Overview
As of mid-2026, AI in forecasting is advancing rapidly, showing a strong ability to generate high-quality predictions across many topics. AI stands out by consistently delivering reliable forecasts at scale, giving it a practical advantage over traditional methods. While matching elite human superforecasters remains a long-term goal, current AI systems are already valuable by performing as well as the collective crowd. This means organizations can now access robust, crowd-level forecasting more easily, solving a major operational challenge and expanding the reach of predictive intelligence without always needing top human experts.