This is, to a large extent, scaremongering. While there are some valid points made in the article, the article fails to differentiate between security problems that can be exploited by trolls or single, untrained individuals, and ones that take a powerful team working on behalf for a government or other such group to exploit. It's the difference between the hospital being defended against your average thief, and being defended against a strike squad of ninjas. Despite this, the article does make good points when it comes to the lack of worry about the problems they found. Even though these vulnerabilities may be over hyped, they are real and the lack of focus on these vulnerabilities is chilling. The real underlying problems for this stem not from an industry that leaves bugs in applications designed for high security, but in the fact that the industry doesn't realize that security needs to be the default, whether or not you see exploits being used.
We wouldn't need to do this. One computer would suffice to generate primes at a pace making cracking computation infeasible. Though, even if we switch primes every few seconds, if somebody wanted to put a few hundred million dollars into cracking a few seconds of messages, they could do it.
The problem is that there needs to be an agreed upon key that each of the parties knows before-hand. But yes, there are definitely viable ways to generate new ones or implement new, safer, standards. Alternatively, a much larger prime can be used. Also, the Diffie-Hellman protocol is a well known one that many many security researchers, programmers, and students have looked at. The flaws are not obvious, as it's initially unclear how "cracking" a large prime would work.
At this point I feel bad for Scott Aaronson. Everytime a crackpot paper comes out (and I'm 97% convinced at this point this is one of them) he takes the time to refute it on his blog.
This isn't exactly Deep Dream's equivalent for video. It takes frames individually (smushing them together), rather than having a unified neural network that takes the entire video at once.
I wouldn't have much hope that its going to be addressed. I've been following another kickstarter scam, https://www.kickstarter.com/projects/181239886/jaesa/descrip..., promising legitimate AI far better than anything that actually exists that appears to be little more than ELIZA plus ads. Its had over 50k USD in funding and is 9 days away from its one year funding anniversery. Obviosly, no real progress towards an AI has been made but the group has managed to keep the scam, recently having claimed to be acquired by on of their backers who is in talks with multiple investors for funding " to the level of a global competitor".
Another example was the Goblins Comics board game, which featured the web comic writer teaming up with a supposed board game maker who than ran away with the money, scamming the board game maker and the comic writer. See more at the second to last post here on the writers continued attempts to help his backers despite being scammed by the board game maker: http://www.goblinscomic.org/the-blog/
Judging by the fact you have to DM "to get send anonymous messages" (as per their profile description on Twitter) I would suggest it's not that hard to track in the event they need to report the end-user to law enforcement.
