Comment by cscurmudgeon

Are there any intrinsic dis/advantages of bidirectional models over causal models for in-context learning? It seems that unidirectional model just have been explored and worked on more.