This parallels its influence on me as well. In particular FCaCA has been a good source to learn how to narrow down problems so they can be tackled without diluting the original essence of the problem. Masterful treatment in my opinion.

I still think Hofstadter's "old school AI" thought processes warrant more attention in today's DL frenzy than they get. His books have been a great source of very interesting questions including style transfer (he dealt with fonts in "Letter Spirit"). It was also the inspiration behind some early experiments using n-nets [1] (shameless plug)

[1] https://blog.imaginea.com/deep-type/