This same thing (i.e., using recurrent neural networks to predict characters (and even words)) was done by Elman in 1990 in a paper called "Finding Structure in Time"[1]. In that paper, Elman goes several steps further and carries out some analysis to show what kind of information the recurrent neural network maintains about the on-going inputs.
It's an excellent read for anyone interested in learning about recurrent neural networks.
It's intriguing how often HP's public cloud offerings are omitted from the conversation. I would think there would be a lot of interest especially because they're a heavyweight yet are using open standards (i.e., OpenStack).
No disclaimer: I don't work for HP (but I have used hpcloud)
But it is important to note, once again, that correlation does not imply causation -- no matter how fancy a scientific tool you are using, either fMRI or genetics.
The best data Kandel cites in support of his argument is the improved treatment outcomes that could be "predicted" due to genetics. However, it would not be too surprising if some behavioral or cognitive measure did just as well at predicting the improved outcome.
For those of us who have never heard of this software, it would be great if somewhere prominent on the webpage it said what GPGMail actually is. Instead the most prominent thing is how many bug fixes and sleepless nights this mystery software required.
Why not stack Morse code horizontally? At least Morse code is somewhat designed to take advantage of the frequency distribution of letters (e.g., e is a single dot.)
It's an excellent read for anyone interested in learning about recurrent neural networks.
[1] http://crl.ucsd.edu/~elman/Papers/fsit.pdf