AI Jam Sessions · Interactive Explorer

Play the phrases. Read the tool-use. See the eval.

An interactive tour of the jam-actions dataset and its preregistered confirmatory evaluation.

Dataset Model ⚡ Live demo Eval write-up Code Collection

▶ Runs entirely in your browser — audio synthesized client-side, no server, no inference cost.

Loading example phrases…

The confirmatory (B-1) evaluation, interactive. Five frozen fine-tunes vs. a fresh prompted baseline, over a preregistered 36-record cohort. Toggle the breakdown; hover the bars.

+0.212
mean Δ, primary condition
29/36
paired wins vs a 24/34 bar
p < 0.0001
exact sign test
10/12
wins on never-trained music

Full write-up with fixed figures & receipts → jam-actions-eval