There's probably people whose main job was to read the code and then communicate it more broadly. This is also cutting edge ML where a ton of code is basically

There's probably people whose main job was to read the code and then communicate it more broadly. This is also cutting edge ML where a ton of code is basically thrown away due to not panning out so possibly the amount that needs to be communicated is fairly small.