DeepMind's AlphaFold2 won the CASP14 protein structure prediction competition by a wide margin and was recognized by the organizers as a solution to the 50-year-old protein folding problem. It scored 92.4 GDT, with predictions averaging 1.6 angstroms of error — about the width of an atom — three times more accurate than the next best system and comparable to experimental methods. The system used an attention-based network that refined a structure graph using evolutionarily related sequences, multiple sequence alignment, and a representation of amino acid residue pairs. It sits outside the generative AI lineage, but was widely reported at the time as a case of AI actually solving a hard scientific problem. Hassabis and Jumper would receive the 2024 Nobel Prize in Chemistry for the work.