There are different ways, some of them involve a lot of math but one way is to start with multiple randomized neural nets which modify inputs you give them to outputs. You then test them, rank them by performance and get rid of the bottom half (doesn’t have to be the bottom half can be another fraction and can be that the lower the score the higher the chance to be removed but random). Then you make copies of the remaining ones with a random change to each and repeat.

This is all automated of course except for setting parameters like how many are removed how many are in each generation etc.

If you want an explanation of a neural net you should look up a video because it’s a lot easier to explain with visuals.

Iirc the one who survived wasn’t even without treatment, there was only a single person who survived after showing symptoms.

I’m too lazy to look for the copypasta about it but basically if you get bitten with rabies and show symptoms you’re already doomed.