Tried out StackGAN's pre-trained bird model: https://imgur.com/a/qgIr1 Can definitely see the limits in the # of species and different colors.
-
-
Replying to @gwern
Hah! I strongly suspected this thing was BS and just regurgitates near-perfect matches. Grumble grumble grumble.
1 reply 0 retweets 1 like -
Replying to @Meaningness
There's definitely weaknesses to its generalization... but it's not a very big dataset either. Not even 'eagles'!
2 replies 0 retweets 0 likes -
Replying to @gwern
My prior is that it will generate garbage when given a meaningfully difficult problem.
1 reply 0 retweets 0 likes
Replying to @Meaningness @gwern
NN people have a culture of not doing control experiments, sensitivity analyses, or anything other than “look! it works! brain woo!”
1:54 PM - 7 Jan 2017
0 replies
0 retweets
2 likes
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.