Fable hand-writes weights of a 100k-parameter legal citation classifier

Legora CTO Jacob Seligsohn says Fable hand-wrote the weights of a 100k-parameter neural network that distinguishes real legal citations from fake ones.

Seligsohn says the hand-written network outperforms a network trained the conventional way. Elvis Saravia, writing on X, calls the result an example of weight-space programming. The reports give no benchmark figures, accuracy numbers or test details, so the performance claim rests on Seligsohn's statement alone.