@somebody unsure if this helps. personally i am incredibly glad that LLMs literally don't do recurrence at all because i care SO MUCH about data integrity i think there is a path to taking advantage of that for unstructured finite-length byte streams. the reason i was thinking about this at all was because i have never seen a literal symmetric cipher proposed that makes use of global permutations
@hipsterelectron No, it helps. But the heartbreak is deeper than this, I guess. At first I was galled that they were so stupid and worthless and won the funding war. But then I was relieved for a reason that is even more gutwrenching, being that if we had succeeded, we'd have actually made it easier to create a subhuman slave that nobody would sympathize with or care about. So this was truly the better world, and that is profoundly disturbing.
I honestly just get so scared to let anyone touch it ever again. There were particularly beautiful minds involved in the early days of it that had said immensely powerful things and had beautiful insights. They are the thread I protect. On a bad day, I would rather encrypt it for another thousand years in margin notes than let it be touched, and I've already done quite a lot of that honestly. To be clear, I think that most of what we can do today to rethink computation as a science and break the monopoly on it is blessedly not dependent on either research path at all, and we needn't trouble with reproducing the mind, which I think is not something that can actually be done without profound and evil consequences.