Daniel Miessler: preference

Transcript context

…And does that practically operate on just a runtime agentic search basis where Claude just decides what it wants to look into and pull stuff into context on its own? Or are you doing some sort of post background batch processing? I've also been quite interested at times in one episode I did was on a system called Hippo Rag, which was taking inspiration from the hippocampus multi-step process where you'd have these, whatever your corpus was. First, you would go through and do entity recognition and deduplification and then create like a graph structure that would have the entities and then the documents in which they appeared. And that way you could rag into it anywhere in natural language, but then see, oh, that connects to these concepts, which connects to these other documents and expand out in a sort of network-based way through the corpus, as opposed to a sort of purely hierarchical approach to retrieving information. That gets pretty complicated, obviously, pretty quick, but it does feel like something like that might be needed relative to maybe this is also just my like lack of confidence in my own ability to organize myself and my thoughts well enough. I certainly do recognize people who are quite different in this regard. But I feel like I need a sort of cross-boundary layer that would probably have to be bash processed in the background to make these connections between all these various disparate things, as opposed to being able to put each one in its proper place such that like Claude intuitively and correctly decides where to go just based on structure. Yeah. This to me is the whole advantage of the scaffolding and like being able to infinitely tweak the scaffolding according to first principles. So because I have the core skill, which is the bootstrap for the entire PI system, because I have that laid out and it gets loaded and it has all the context of what we're trying to do and everything, it gets also the architecture of the system, including the memory system. Now, all this stuff that you're talking about doing with like scripts and stuff, like that is the clog code hook system. The clog code hook system is extraordinary. So, I have, I think, right now, 12 hooks that are active. And this is, I've got a whole bunch for user prompt submit. So, there's security checks in there, there's sentiment analysis checks in there. It's actually routing throughout the PI system according to what I'm trying to do based on the sentiment analysis, which uses Haiku. So, I have a custom inference tool, which has three levels of inference: fast, standard, and smart, which is Haiku Sonnets and Opus. And so, the entire system is using this to like self-route. Now, the memory system and all those sentiment analyses and all the artifacts of keep in mind, this is it's fully archiving. Cloud code does this naturally. Every prompt I send, every tool use that it runs, every output of the tool use, this is all recorded. It's all there raw for us to analyze. So, I am taking that and putting it inside of this memory structure, and I'm overlaying on top of it sentiment analysis. This is all being done dynamically. I'm not seeing anything, it's all just handled automatically due to hooks. So, hooks are constantly adding the sentiment layer of how good the algorithm is doing, how good the Pi system is doing overall. So, at any point in time, I could say, What upgrades have we made to the system? How have they gone? How has our performance been going in the last month? And Pi will come back. Kai, in my case, will come back and say, Yeah, it seems we tried this, that didn't work. We uninstalled that. We went back and we went in another direction. And currently, we're doing this, and you seem much happier with this. So, this seems like a direction to go. Do you want to do any more work on that? So, and that's all just operating on raw logs. There's not like a summarization level or some sort of because that sounds like just like a ton of content for it to wade through.…

Stored transcript either side of the excerpt. The highlighted words are the published quote; the surrounding text is unedited source, never generated.