I never posted about it, but I spent spare time at Consonance this year implementing the new neural net encoding. I then did nothing with it until the flight back from Confluence yesterday, fixing the one silly bug that was left. And, using it, I finally got a neural net to evolve. Barely. It can reliably and easily find a solution for the most trivial task possible: echo all 1s from the input to the output. Anything more complicated and it fails. But since this project is at least 6 years old, I'm pretty happy to get any results at all. (And now I'll probably forget about it for another year or two.)