4 ms·
Very good paper, was thinking about similar steganography technique, but using Markov chain instead of GPT. GPT output is obviously more realistic, hence better
by coolspot 5y ago
Very good paper, was thinking about similar steganography technique, but using Markov chain instead of GPT. GPT output is obviously more realistic, hence better covering the presence of hidden communication.
In the colab, I can encode a message, but I don’t see an easy way to decode a message, e.g.:
The technical challenges of embedding oulogramming messages in real world communication settings is discussed. Its application to centralized powers of censorship is defined
=> ppl: 44.14, kl: 0.012, words/bit: 0.26, bits/word: 3.90, entropy: 8.65