(10:09)System: Initiating private chat with active user ItsDuckKingKarl.
(10:12)SonjaOfTheMoon: Look at AgentPlanningManager.ag. You see that giant mass of compute nodes? (10:16)SonjaOfTheMoon: It keeps getting passed garbage data. It's not intense enough to actually do anything, but it does force the simulation to deal with it, and with a structure this complicated that's a significant waste. (10:17)ItsDuckKingKarl: Alicia and I keep patching it with filters on the inputs, there's probably more you can do with that (10:17)ItsDuckKingKarl: Have you tried running a correlation analysis on input intensity vs no-op node activation? (10:18)SonjaOfTheMoon: Yes. I see some options with filter nodes, but frankly the idea irks me. There's got to be some way of tracking down the bad input in the first place... (10:21)ItsDuckKingKarl: @hena is fundamentally different from those kinds of imperative languages. It's a mixture of Church's lambda calculus and some of Kurzweil's discoveries about networked intelligences, designed to take advantage of emergent physical properties of contemporary chip miniaturization. You can't think about it as if it were deterministic (10:23)SonjaOfTheMoon: I know that. I mean, I don't fully understand the mathematics behind how those three things can coexist, but I know how to reason through an @hena network. (10:24)ItsDuckKingKarl: It doesn't really make sense to talk about where any particular @hena data comes from, because so much of the program is emergent properties of the node graph itself (10:24)SonjaOfTheMoon: Charles, you don't need to start quite this basic. (10:25)ItsDuckKingKarl: If had to load-balance two compute nodes θ and φ with some common dependency ψ, how would you do it? (10:26)SonjaOfTheMoon: Uh, pipe noise through a balancer node connected to both? (10:27)SonjaOfTheMoon: ( @balance ( ( @noise ( ) ) ( θ φ ) ) ), perhaps? (10:27)ItsDuckKingKarl: You wouldn't just put the balancer on between the nodes and the dependency? (10:27)SonjaOfTheMoon: Wouldn't that create an implicit dependency between ψ and the balancer? You'd have a cyclic dependency graph... (10:27)ItsDuckKingKarl: Right, but noise nodes are already implicitly dependant on each other due to global RNG state (10:27)SonjaOfTheMoon: Only as an implementation detail. You're testing me. (10:29)ItsDuckKingKarl: Those same low-meaning blips in the planning manager set up vacuum perturbations that encourage tunneling effects in subsequent activations of the same system (10:30)ItsDuckKingKarl: So anything that just shuts them off will slow down the planning manager overall (10:30)ItsDuckKingKarl: You want to make them less frequent, except right before something actually uses it for real (10:34)SonjaOfTheMoon: What, an @hena extension to create systems with specific physical side-effects? (10:34)ItsDuckKingKarl: Google's demonstrated remote information exfiltration through a similar technique, it's not that much of a stretch (10:35)SonjaOfTheMoon: I heard about that... They read data off a hard disk that their scanner had no direct connection to, right? (10:66)ItsDuckKingKarl: It was an old drive chosen specifically for being able to break into (10:37)ItsDuckKingKarl: If you want to try optimizing the planning manager, that's where you could start