Fluid Concepts and Creative Analogies is a wonderful book. I wouldn't call the systems they build upon (such as lMetacat [1] and Tabletop [2]) "machine learning"...
I'd also like to see FARGonauts / Center for Research on Concepts and Cognition style software implementations around... I have run with success this python implementation of Copycat:
I implemented it in Python (while reading from that LISP version, a Java version, and Melanie Mitchell's "Analogy-Making as Perception"): https://github.com/jalanb/co.py.cat
Its maybe a matter of taste but I find his other work to be rather pretentious, although geb is entertaining and somewhat useful (I half liked le ton beau de marot).
I really think k there's something useful.in fluid concepts that maybe can help with low data learning. I think geoffrey Hinton hints at it in his lecture about trying to do pca on learned neural states to find manyfold correlations which in high dimensional states are stronger to be nonrandom.
Boelter said: “[Some] might say that this vulnerability could only be abused to snoop on ‘single’ targeted messages, not entire conversations. This is not true if you consider that the WhatsApp server can just forward messages without sending the ‘message was received by recipient’ notification (or the double tick), which users might not notice. Using the retransmission vulnerability, the WhatsApp server can then later get a transcript of the whole conversation, not just a single message.”
I guess what they're suggesting is to compromise the server in such a way that it does not send "delivered" receipts for any messages anymore (even though it actually delivers them to Bob, and Bob answers, and a "normal" conversation ensues).
Then, at some point later, Eve on the compromised server could send a "oops, here's a new key, send everything undelivered again" message. Then, the client, as it is now, would just re-encrypt and re-send all those messages it deems undelivered so far (and then pop up the "key changed" message, if you had requested it in the settings).
You'd recognise the attack by seeing only single ticks on messages, even if Bob had seen them and answered.
People underestimate a lot the power of "useless" features, if well designed, to attract new users.
Telegram's Stickers are probably one of the features that most people feel other messengers lack to make people enjoy using it. Lots of friends are actually communicating daily through GIF's.
Google is supposedly hiding a direct competitor to one of its main product from its search engine, the main entry of the Internet for millions of people that aren't very proficient with computers.
Google's product, Gmail offer considerably lower security and privacy than this competitor and using it instead of ProtonMail may very realistically get someone (a journalist, whistleblower, LGBT activist...) in deep trouble or even killed in lots of places and contexts.
I think this is a pretty evil move. Not only for this but what it represents when, again, millions of people were educated for years to enter the Internet through this search engine and don't have any clue that it may not only tamper with the order of the results but also hide websites.
If you are using ProtonMail for anything critical you're doing it wrong. It might be more convenient than other encrypted mail solutions, but it's in-browser cryptography with all the security issues that entails.
Sure it is not the best tool -- but depending on what means "critical" in your context it provides great security. Not against the NSA targeted attacks, sure, but if you're not so concerned about your adversary using exploits and more sophisticated tools (versus just demanding data from the provider or something like that) ProtonMail goes a long way.
Anyway it is, for most cases I can think of, much better than Gmail.
Interesting note: The author is part of the rotor browser fork that is going no where so far. Doesn't look like the reported issue has been fixed there. In fact, no commits since before this blog post."
