Very glad to have come across these notes. For quite some time now I had been actively attempting several different methods to overcome the very same problem. The closest to my need has so far been the 'capsule' approach. Quick explanation would be definite repititive systematic encryption to reduce text volume down to less than 10% at every instances when the cumulative conversation history has crossed some specific threshold, say 10 million characters total (including the capsule of N-1th round) and then automatically erasing the long version for every iteration and print the capsule in the core memory, not overwriting the older capsules though, very important to call that out, for itself to have a reference point of cross validation, whether the decryption was accurate.
Very glad to have come across these notes. For quite some time now I had been actively attempting several different methods to overcome the very same problem. The closest to my need has so far been the 'capsule' approach. Quick explanation would be definite repititive systematic encryption to reduce text volume down to less than 10% at every instances when the cumulative conversation history has crossed some specific threshold, say 10 million characters total (including the capsule of N-1th round) and then automatically erasing the long version for every iteration and print the capsule in the core memory, not overwriting the older capsules though, very important to call that out, for itself to have a reference point of cross validation, whether the decryption was accurate.
Let me know what you think of this approach.