About time Jonas made the front page. He's been on HN a while and freely answers questions, both here and on his twitch stream (http://www.twitch.tv/317070).
Sure, author here, ask me anything!
I love it! Looks more interesting than most stuff I've seen people draw on acid. No surprise there were a lot of drug users interested in it lol. More professionally, great that we have so many ways of debugging them now which are visual and hence easier for developers to understand.
Rule 34? I'm sure someone has trained this on porn already? That'd make for some messed up images.
We haven't seen porn on there yet. The most used word is 'nipple' (which in the USA would be a pacifier) leading to vast disappointment.
Best for porn would probably be bathtub, brassiere or bikini (or a mixup of the previous).
I don't think there are dirty pictures in the imagenet-1000 dataset, so the network doesn't know of porn. But there are 1.3 million images in the dataset, so we don't know for sure.
Amazing concept, and now there is bot acknowledging people petitions.
That has been there since the start :) it randomly picks from the chat suggestions every minute and acknowledges the one it picked. I'm one of Jonas's co authors, feel free to ask me anything as well!
Are there similar results for audio training data ?
Not to my knowledge. For speech it should be sort of doable to get enough data to train a large net and then do inceptionism-style stuff with it. For music, gathering enough training data is a huge challenge due to licensing issues, among other things. My main research topic is deep learning for music, so I'd love to try it sometime.