LLMs

“All of this research happened in Google. And LLMs were actually, or these language models were actually built, you know, with one purpose, which is to make search better, to make Google search better.”

“And actually, like, it's always been part of Glean. So even on day one our product, we're using LLMs.”

“You build robotic process automation with intelligence. Those solutions are the most scalable because if you just have one piece it won't scale because if we just build intelligence, it's a, Recommendative engine, so people don't see an uplift or a value right…”

“I think at some point, we'll get to a situation where companies are not AI companies the way companies are not, like, app companies anymore, right? Like, there was a time when everybody's like, oh, we got to make an app. We're an app company. We're going to ma…”

“And where I think it is most useful is inside enterprises. So both for their internal use and for building stuff for pro for customers. But I think that's where it adds the most value.”

“The hallucination rates were significantly worse in 2019 than they are today. You know, I internally joke, I think the hallucination rates have gone down, but the ones that happen are so hard to detect that they can, you can get away with them.”

“The whole point of all these LLMs is to understand natural language, so, like, you better believe that they're all trying to Make prompt engineering effectively useless, because they're just going to understand what you want when you say it. So, like, there's …”

“Everyone just so focused on text, like, what was going on with LLMs, that no one realized that, oh, of course, like, other mediums are gonna be way more human and popular than the text medium.”

“LLMs are just, like, much more farther along, and they have, like, a very well-defined, like, path, and it's really just about scaling them at this point, and, like, You know, working on reasoning, improving general intelligence. I think for, like, media gener…”

“This number started increasing, like, sometime in, like, sort of the second half of the last year when, like, models just became better, and, like, you could actually build real-world agents out there. That kind of worked, and a lot of, like, startups coming i…”

“I feel like this is a 20, 3040 year journey, and there'll be new jobs for humans, because simply because we evolve faster, and we need to create history for the LLMs to copy from, you know, because they are basically caricatures, and of course, they do have em…”

“Like, you know, before, if you imagine like a year or two ago, you had to put a lot of effort into very bespoke ragstacks and routing, but now LLMs are getting more capable of doing more agentic behavior. As a result, you have to change your architecture. Othe…”

“And you have to take that and convert that into those hundred bespoke workflows. Which was economically unfeasible before LLMs, but with LLMs, we saw that we were actually able to go deep into the workflows, and as a result, provide more value for the firms, a…”

“There'll be a set of deterministic rules too, that humans for a while will have to like layer on top of the LLMs, everything that we're doing, because I don't think it's going to be fully agentic forever.”

“So I think we build the product from an architectural standpoint a lot different, where when complexity increases, it's not that we increase the prompt size. We increase the number of components of AI components of LLMs. That we use. So essentially when comple…”