@kragen@nerdculture.de
As far as I know you also can not guarantee that it does what you want it to do (or anything at all), outside of statistics, so if it's slightly different but still covers the same percentage of cases does it really matter for software freedom? I guess it might matter for forensics, but that's beside the point.
Can you just randomly stumble upon a substantially better neural network? There has got to be a method even to that madness, it can't just be purely random brute force. I would not accept our AI overlord if it's nicknamed "jackpot"!
@namark @kragen @cwebber @phoe @ivan
I feel that we've veered way off track of my original point, which was just that we can't simply argue that the solution to shipping software around is always going to be to ship source code. It's a nice idea in theory, but it bangs up against practical limitations, thus we'll eventually need to ship binaries around, and once we do that, we need to think of what the implications of that are in an AGPL-ed world.
I see, I guess I'm missing the context where "just ship the source" was suggested as solution to shipping software. We have been shipping binaries around already, along with source code. Could do it better of course.
Otherwise, my (possibly off topic as well?) point was, that in my opinion, in practice, under AGPL, as a node in a p2p network, you don't have to provide binaries or config files with sensitive information in them, unless this sensitive information is miraculously central to the source code of the program and its effects can be observed by others on the network.
