Stanford prof: Google overstated AlphaGenome claims, but calling it 'AI slop' is nonsense

anshulkundaje · x · 2026-09-18

Stanford professor Anshul Kundaje weighs in on Google's AlphaGenome marketing and its evaluation:

Marketing claims (e.g. Sundar Pichai's "we mapped effects of 9 billion variants") are overstated: the model predicted effects of 9 billion SNVs across a few thousand cell contexts, providing an aggregate variant prioritization score.
But he strongly rejects calling such variant effect scores—or outputs of sequence-to-omics models generally—"AI slop".
He judges the paper's evaluation more than sufficient: a useful variant prioritization score and a useful lookup resource.

More from Models