momodomo

Sounds like a good way to solve a corner case in speech recognition: speech to text for names. In reverse, it could also improve speech synthesis for personal robots that would speak the names of their users.

I also suspect that Palantir is involved.

Similarly, will it float so low that it could be shot down with a gun or missile?

To avoid these two risks (storm and attack), I imagine it'd have to be at an altitude above weather currents and beyond shooting range... or be able to stabilize during a storm and be bulletproof.

If I wanted to compress 100:1, how would I select which words would be the "heads" of my chains?

Choose every 100 words in the dictionary?

Is there a risk of not getting complete coverage? Example: I keep applying the one-way function to the ciphertext, but the result is never an endpoint of one of my precomputed chains.