Nice article, lots of good tidbits and insights. The thoughts on indirect encoding seem like an especially interesting area of research. I actually just finished up a project that used a genetic algorithm to tune some of the hyperparameters of an RNN [1]. I didn't evolve the architecture except which cell to use (LSTM or GRU), but it wouldn't be much of an extension to the proof of concept I coded. Thanks for sharing.

[1] http://cole-maclean.github.io/blog/Evolving%20the%20StarCraf...