Claude's mistakes stem from bad instructions, not weak AI

Original: Andrej Karpathy: "90% of Claude's mistakes come from missing context, not a weak model."

Source: x.com ↗