Recently, there has been increased focus on performing approximate Bayesian inference, often using Integrated Nested Laplace Approximations (INLA), particularly with large-scale problems. While this may be computationally more attractive alternative to methods such as Markov Chain Monte Carlo, there are still issues that need to be addressed. During this talk I will explain what INLA is, why it's important and some of the interesting challenges in Numerical Linear Algebra that will need addressing in the future.