Slides for today’s talk at the MSRI Workshop giving an overview over our recent works that prove how noisy interpolation can have close to 1/n rates with the right inductive bias but exhibits peculiar phenomena in the context of robustness.