Pareto smoothed importance sampling leave-one-out cross-validation (LOO-CV) is a method that's used to approximate the LOO-CV results but without actually performing the K iterations. This is not an information criteria, but in practice provides results that are very similar to WAIC, and under certain general conditions, both WAIC and LOO converge asymptotically. Without going into too much detail, the main idea is that it is possible to approximate LOO-CV by re-weighting the likelihoods appropriately. This can be done using a very well-known and useful technique in statistics known as importance sampling. The problem is that the results are unstable. To fix this instability issue, a new method was introduced. This method uses a technique known as Pareto smoothed importance sampling (PSIS), which can be used to compute more reliable estimates of LOO. The interpretation is similar to AIC and WAIC; the lower the value, the higher the estimated predictive accuracy of the model. Thus, we prefer models with lower values.