Legora's CTO says Claude Fable hand-wrote the weights of a 100K-parameter legal citation checker that beat trained versions

Jacob Lauritzen, CTO of legal AI company Legora, says he got Anthropic's Claude Fable to hand-write the weights of a roughly 100,000-parameter network that tells real legal citations from fake ones, one-shotting a model he says is better than training it. In his results, the hand-written network saw zero labelled examples, uses 3,793 non-zero parameters and scored 97.8% test accuracy (98.9% after one revision with 5,062 labels), against 76.5% to 94.6% for conventionally trained 107,745-parameter versions given 512 to 8,192 labels. His own, unverified figures. It follows MIT CSAIL's Weight-Space Programming paper showing frontier models can read and edit network weights directly.