Most inferential methods for high dimensional regression focus on estimating the linear coefficient beta. This focus is the result of trying to transfer low dimensional thinking into the high dimensional world. Inference for linear coefficients is not meaningful unless we impose very strong, untestable assumptions. I'll talk about assumption-free methods for inference in high dimensional regression. The only assumption we make is exchangeability. We do not assume linearity, Normality, constant variance or restricted eigenvalue conditions.