Alessio Fanelli: belief

Transcript context

…I agree, it's challenging there. Okay, so the last piece I'll mention is briefly, ML Ops is still real. I think LLM Ops or whatever you call this, AI Engineer Ops, the Ops layer on top of the LLM layer might follow the same evolution path as the ML Ops layer. And so, the most impressive thing I've seen from the ML Ops layer is from Apple. When they announced Apple Intelligence, they also announced Teleria, which is their internal ML Ops tool, where you can profile the performance of each layer of a transformer. And you can A-B test like 100 different variations of different quantizations and stuff and pick the best performance. And I could see a straight line from there to like, okay, I want this, but for my AI Engineering Ops, like, I want this level of clarity on like what I do. And there's a lot of internal engineering within these big companies who take their ML training very seriously. And I see that also happening for AI Engineering as well. And let's briefly talk about RAG and context caching maybe, unless you have other like LLM OS stuff that you're excited about. LLM OS stuff I'm excited about. No, I think that's really a lot of it. It's like move beyond being observability or like help for like making the prompt call and like actually being an LLM OS, you know? I think today it's mostly like LLM Rails, you know? Like there's no OS, but I think like actually helping people build things. That's why, you know, if you look at XLA-A2B, it's like, that's the OS, you know? Those are kind of like the OS primitives that you need around it. Yeah. Okay. So I'll mention a couple of things then. One layer I've been excited about publicly, but I haven't talked about it on this podcast is memory databases, memory layers on top of vector databases. The vogue thing of last year was vector databases, right? Everybody had a vector database company. And I think the insight is that vector databases are too low level. Like they're not very useful out of the box. They do cosine similarity matching and retrieval, and that's about it. We'll briefly maybe mention here BM42, which was this whole debate between Vespa and who else? Quadrants. Quadrants and I think a couple other companies also chipped in, but it was mainly a very, very public and ugly theater battle between benchmarking for databases. And the history of benchmarking for databases goes as far back as Larry Ellison and Oracle and all that. It's just very cute to see it happening in the vector database space. Some things don't change. But on top of that, I think one of the reasons I put vector databases inside of these wars is in order to grow, the vector databases have to become more frameworks. In order to grow, the ops companies have to become more frameworks, right? And then the framework companies have to become ops companies, which is what LangChain is. So one element of the vector databases growing, I've been looking for what the next direction of vector databases growing is, is memory. Long conversation memory. I have on me this B, which is one of the personal AI wearables. I'm also getting the Limitless personal AI wearable, which is like, I just wanted to record my whole conversation and just repeat back to me or let me find, augment my memory. I'm sure Character AI has some version of this. Like everyone has conversation memory that is different from factual memory. And right now, vector database is very oriented towards factual memory, document retrieval, knowledge-based retrieval, but it's not the same thing as conversation retrieval, where I need to know what I've said to you, what I said to you yesterday, what I said to you a year ago, three years ago. And there's a different nature of retrieval, right? So there's a, at the conference that we ran, graph rag was a lot of focus for people, the marriage of knowledge graphs and rag. I think that this is commonly a trap in ML that people are like, they discover that graphs are a thing for the first time. They're like, oh yeah, everything's a graph. Like the future is graphs and then nothing happens. Very, very common. This happened like three, four times in the industries past as well. But maybe this time is different. Maybe. Unless. Unless. Unless. So, this is a fun, this is why I'm not an investor. Like you have to get the time. This time is different because no ideas are really truly new, but sometimes this time is different. Maybe.…

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