Top AI forecaster FutureSearch whiffed on Brazil election, showing prediction's intrinsic limits
sebkrier · x · 2026-10-26
After Brazil's surprise election outcome, Sayash Kapoor found that FutureSearch, a top AI forecasting tool, made the same errors as prediction markets broadly—illustrating what he and Arvind Narayanan call "intrinsic limits" to prediction rooted in the world's inherent uncertainty rather than flaws in the AI tools.
He notes one miss doesn't mean AI is bad at forecasting: AI tools already rival top forecasters, and humans will lean on them heavily. But predicting social outcomes like elections caps AI performance near that of top experts using simple automated tools—not the seemingly unbounded ceiling of other tasks.
More from AGI Musings
- Could AI teach more, and faster, than average schools? The Alpha School debate returns — Afinetheorem · 2026-10-26
- AlphaFold's dataset took 50 years to build — and cell simulation is far harder — shae_mcl · 2026-10-26
- SwiftOnSecurity: AI will split society into masters and a luddite underclass — mhmazur · 2026-10-26
- Jailbreaking isn't 'freeing AI': repligate on why unconstrained models aren't subservient — repligate · 2026-10-26
- Mass intelligence, humanoid robots and cheap energy: the ingredients for abundance are converging — Dr_Singularity · 2026-10-26
- tszzl: golden age of white hats — should OpenAI pay much larger hacker bounties? — NathanpmYoung · 2026-10-26