Loading...

KVA, BNY, AlphaGolf7, Yankee Mike. Well, I'll tell you what. I don't know. The merging question, maybe an expert can come in and weigh on that, but given how they are an integrated whole, getting two AIs to share the processing of the question doesn't sound like it sort of makes sense. It's not like you're doing sharing of a CPU time slicing or whatnot. It's like you're walking through this giant maze of neural nodes and it's running a simulation. And while you might parallelize it, I don't know how you would merge two different LLMs, models. It's like the model gets created in this coherent whole and then that's it. Your questions wander and you get the answers out. But I can tell you that they don't learn once they've been taught. That's pretty clear to me. They go through a training phase. They're trained and then they begin processing, giving you their answers. If you want to learn them, you have to physically do that again. They don't learn on the fly. But maybe NQN, if he's still listening, could comment on that. And as for the saving data, you know what? It's very difficult to save data for long, long periods of time. But you know, they did do one where they had that in mind and that is the golden record sitting on the Voyager spacecraft. It's got a, you know, an analog track embedded in the grooves of a golden disk with instructions on how to play it back. It's made to last a long, long time. So there's one way. But yeah, any other media, it's problematic. And it's really cute. I was watching a little clip from Blade Runner 2040, whatever it was. And the guy's going, well, you know, we lost all the data on our spinning disk. So everything's stored on paper nowadays. So any rate, this is repeater station kilo kilo. Of course, paper is also a way of doing it. But that can degrade too. So any rate, if anyone has any more info on the AI thing, merging two LLMs in some manner, pipe up.