This is terrifying. If this, and similar ideas, gain adoption, this will undoubtedly lead to the next financial crisis.
From my understanding, JPMC wants to issue a new coin and make the market for that coin by buying the coin for $1. Then, they plan to use this coin for it's normal banking tasks, such as lending money, except they would use JPM Coins instead of dollars. By using this coin, they would be able to skirt their fractional reserve requirements. Of course, this coin only has value because of trust in JPMC. If this trust or breached, or if external economic events cause a large amount of JPM Coin to be sold, then this will effectively be a run on JPMC. Given JPMC's size and importance in the marketplace, the aftermath would not be pretty.
This isn't that surprising. I think the reason for this is that, even though the model is linear, the space of n-grams is so large that there usually is a line that separates any two classes.
The purpose of this work isn't to generate realistic NSFW images. The purpose is to visually represent what a neural network "looks for" when classifying NSFW content. It is very interesting that these generate images strike the viewer as grotesque even though there isn't anything specific in the images that could be called vulgar. This indicates there is some similarity between how a neural network and our brains classify things.
You should definitely CDN your data. A majority of a client-side latency is spent round-tripping to the server. Using a CDN greatly reduces this time.
I don't know your workload, but 60ms doesn't seem that fast. Don't get me wrong, it's pretty good, but for a read-heavy load it seems like this could be optimized to be sub-10 ms at the 99th percentile.