← @bgSponge

Artificial intelligence now beats some of the best human forecasters

Summary

An AI won the Summer 2026 Metaculus Cup for the first time — AIs also took second and fifth, humans third and fourth — per an Economist report citing Forecasting Research Institute analysis.

  1. Metaculus Cup milestone.

    An AI won the seasonal cup for the first time; AIs also took second and fifth, leaving third and fourth to humans across hundreds of September-resolved questions.

    For the first time an AI won the seasonal Metaculus Cup, a proving ground for forecasters of a wide variety of events.

    The human forecasters who enter Metaculus’s competitions are not necessarily the world’s best.

  2. LLM forecasters, not narrow models.

    Unlike weather-style specialized systems, these bots use general LLMs to read news and data broadly and explain reasoning—useful when persuading decision-makers.

  3. Startup tricks on top of scale.

    Multi-model investigation and debate, paywalled esoteric datasets (Mantic), pundit track-record scoring (Preseen), and memory wipe for counterfactual re-runs sit atop frontier LLM scale.

  4. Solo tinkerer can still win.

    Jeffrey Liang’s bot beat better-funded startups with under 150 hours and a couple thousand dollars of compute, and leads an AI-only Metaculus contest.

  5. Limits and price collapse.

    Four-month cups and non-elite human fields bound the claim; multi-year forecasting may still favor humans. Superforecaster work can cost over $10,000 and a week; FutureSearch sells minutes and a few dollars.

    suggested AI systems have now reached parity with the superforecasters on an evolving set of forecasting questions.

    A forecast from human superforecasters can cost more than $10,000 and take a week.

Read the original ↗Save to your Sponge

Summarized with Sponge. Quotes are the author’s own words.

More from @bg

All readings →