Finding Sparse Linear Connections between Features in LLMs — AI Alignment Forum