Thought Email List: unsolvable puzzle fun Subject: Re: reversing xor with random data was: Re: reversing simple hashes I'm noticing there's likely a cryptographic vulnerability in things stimulated by the lack of use of appropriate transfer learning in machine learning. Could validate sharing relevant transfer learning, although I hear there's a transfer model out there. But maybe I'm just insane? I'm probably just insane. It's hard to never try your ideas out for a decade. [the idea is to train models that learn to transfer well by conditioning their success on data they have not seen throughout the whole training, might need metalearning, demonstrates my lack of understanding of normal practices] I'm thinking about it more and realizing people would just pretrain large models that predict random number generators and not worry about it. Comparable --