Hacker News
Derek Lowe's Verdict on AI Drug Discovery: Clinical Evidence Is 'Disappointingly Limited' and the Two Program-Killers Remain Out of Reach
Lowe's In the Pipeline post from 2026-08-10 resurfaced on HN this week at 147 points and 81 comments — the community validation is the new part, not the piece. His assessment is that the volume of AI drug discovery papers, talks, and press releases makes it genuinely hard to determine what the applications are or whether they help, and that evidence of real clinical impact so far is disappointingly limited. He holds the position he has held since 2025: better target selection and human toxicity prediction kill most programs and neither is within AI's reach. Useful as a calibration point for how a domain expert reads AI capability claims in a field with hard, slow, expensive ground truth.
↳ Follow the thread