Thought Email List: unsolvable puzzle fun To: confident naive puzzler From: shaky naive puzzler Subject: Re: reversing xor with random data was: Re: reversing simple hashes
.... or a _simplistically biased random process_ like 0 90% 1 10% ... any of these would be reasonable. Maybe the simplistic bias? To keep things kinda noisy but also useful?
Then, um, uh, for the user text, we could use GPT2 ? Or .... I dunno! How about GPT2? This sounds like it would work great!
I don't really attend to cryptographic research to that degree or any more, but I don't think I've been exposed to something that engages a one time pad, so I think it could be fun to try to implement it. If you did the 10% vs 90% then you can form probabilities around each byte happening independent of user text, conditioned on user text, like in the recent probability thoughts. Then, GPT2 outputs its own probabilities. The most probable result could be calculated from those values.