NARKOZ
[Enhancement] Turing test failed on significant other for SMS responses
I suggest making more believable responses that show some affection and more variety will make the script more useful. @popaulina was not pleased with me attempting to put this into practice. She has volunteered her time for the cause (we can use her help for testing).
I suggest ideally some sort of language modeling for these excuses (e.g Markov sentence generator), or at the very least marking excuses that were sent recently to be excluded for the next 6 months or longer if you've recently fucked something up in your relationship.
Thoughts?
Recommend using a significantly large enough corpus to train a lightweight RNN ...preferably full of an inordinate amount of tech jargon but with a limited output vector...i.e. no more than a few words based on a string constant with about 4 times the length and twice the kolomorogrov complexity of the random output appended...backed of course by the large jargon training set. Also need some kind of callback or relay to send the excuse back to the user for synchronization purposes--in addition to a failsafe default can throw randomized garbage into the email under certain circumstances allowing the user access to plausible denials such as "I've been hacked" or "coworker prank"...now how that for a bike shed :P
