Not a machine that cures things

AlphaFold's achievement has a precise boundary.

AlphaFold predicts the shape a protein sequence is likely to fold into. That is the claim. It is a large claim, and it is the whole claim.

It did not discover or test a drug. A predicted structure may help researchers propose or rank candidate molecules. Candidates still need laboratory studies and, if they progress, clinical trials that test safety and effectiveness in people. Many candidates fail somewhere along that process.

A model can propose, rank, predict, and narrow. Those are prediction tasks. Its output does not establish that a treatment works in a living body. Experiments supply that evidence.

AlphaFold also has limits as a structure predictor. It often produces one structure when a protein can take several forms, does not include many or other non-protein parts, and is not validated as a general predictor of what a will do. The database reports confidence so researchers can see where a prediction needs more caution.