[tommiegannert]

On a meta-note: I've thought for many years that the next step for cryptography is steganography. (I.e. true ones, not disk encryption that has a hidden partition, and unlocking one will make it obvious there's another...)

I did not expect machine learning to be the platform that paved the way for it. I thought we'd embed one file in an unrelated other file, and that some HMAC/AEAD algorithm would allow you to quickly know if a file has data hidden data you can decrypt. Both files being useful in their own right. Instead, we'll just flood the tubes with variations of the original, hiding the original in the crowd.

Ugh. The stuff we spend energy on producing.