AlphaFold 2 changes the protein-structure problem at CASP14
What happened: DeepMind reports accuracy competitive with experimental structures in the blind CASP assessment.
Why it matters: A decades-old scientific bottleneck becomes a tractable learned-inference problem.