In the 2nd stage, this contextualized function is used to resolve the conditional subsequent pixel prediction task. The noticed two phase efficiency of our linear probes is paying homage to another unsupervised neural net, the bottleneck autoencoder, which happens to be manually designed to make sure that capabilities in the middle are used.Specif