Great article, thanks for sharing! I’d add that the reason that GP inference scales with the cube of the number of data points is that technically GP works by inverting the covariance matrix, which has cubic computational complexity.