Fly Language Model built on real connectome 'predicts' House races with 89% accuracy

nrmarda · x · 2026-09-13

A researcher has unveiled the first-ever 'Fly Language Model,' incorporating the real neural connectome of a fruit fly into an AI system. Someone immediately took it further: feeding congressional district partisanship and other variables into the fly's sugar and bitter taste neurons, letting the insect steer its body to 'predict' races — achieving 89% accuracy on U.S. House races.

The viral framing: 'this is my quant' — an insect brain outperforming what sounds like it should require far fancier election models.

Original post →

More from Fun

Fun channel →