LLMs

“Yes, there's AI, and, like, AI's gonna end the world, but, like, that's actually, I think, at this point, like, feeling more fringy. Like, people are excited about LLMs in various ways, but it's not like a, it's not as cultural a movement as it was, like, a wh…”

“There really are two use cases. One is Dave's point, which is summarization, and the other is Brisk's case, which is data theft, right, and data laundering. Like, I agree those are the only use cases”

“Summarization is the only great Use case of LLMs currently today, because in the case of summarization, you have far less hallucination issues.”

“Like, there's no way that doesn't happen over time. It's too appetizing. The data asset is too, Like, the centralization of that much data, of that much value, like, you cannot keep your hands off of it.”

“But I just think this stuff is all gonna, the core stuff will be all very commodity, whether it's decentralized or not. And commodity things end up getting competed mostly to zero.”

“What's really happening, I think, with a lot of this AI stuff, is the public web is dead. People are not going to post shit publicly, right? Because why would you give away all the value, get nothing in return?”

“We've kind of come out publicly and said, hey, for all the creative models, we're going to build our own, because we feel like we have the data sets and the experience on that, but on the LLM side, we're partnering with a number of these folks, and we are just…”

“And I think I'm, as you know, extremely skeptical in the, like, this is some sort of like, this is asymptoting. Like we know what this looks like and it's going to be useful and like, it's good. I don't think there's some like major next breakthrough.”

“I think LLMs like the really cool, very obvious applications. There'll be a lot of money made. I don't think it's like the second coming of anything. I just think it's like, Cool business cloud, two point O stuff.”

“Science has not yet proven that you can create higher quality LLMs using AI generated data.”

“And I think that that ethos is going to play out in this realm over the next 10 years as powerfully as like the centralized ones do.”

“And so all these licensing deals the LLMs are trying to do isn't for training per se. These things have been trained 12 months ago, but it's this layer which is really important, they believe, to sort of the fact ecosystem.”

“It's very hard to compare what's going on to the internet wave because LLMs are not networks, they're computers, and they're very simple. It's like you put in A bunch of things happen inside a box and output comes out and that's a computer, not a network.”

“LLMs are great. They're a really interesting breakthrough in how AI technology works, but they are not sentient even remotely. And I think that this thing you're talking about, Britt, related to it telling someone to jump off the Golden Gate Bridge, this is ex…”

“If you take a huge amount of data and play guess the next word, you can do amazing shit with translating languages, and it's super valuable. You can, for sure, make more compelling ads. Like, you can make, Fake porn that, like, is really titillating. Like, you…”

“Large LLMs will belong to the large players, players who can run on train on very large clusters, people who can pay for proprietary content, whether it's with Reddit or somebody else.”

“Obviously from an open AI perspective, it clearly establishes where open AI is in the competition for the best LLMs. Obviously many people wanted that business. And many things were evaluated, and so it's really rewarding to see where Apple thinks the best AI …”

“Ad creative targeting, like these are the places that LLMs have by far the most leverage in the moment we're in.”

“The way I think it's similar to social media is that you need a new ad format that doesn't exist yet.”

“I don't think the end is here. Wrangling LLMs is the thing going to be a thing for a very long time, but I think the way that we do it, There's going to be a lot of UI.”

“And I think that's, like, kind of the reality of all these LLMs to remember is that they don't actually tell you truth. They just tell you what you want to hear.”

“OpenAI put out a new study about software engineering tasks and whether or not LLMs can pick up random software engineering tasks off Upwork, and what percentage of them can they complete? And it's only something like 40%, and crucially, LLMs can fix bugs, but…”

“It's just like, eh, and like, I just don't see any pull-away differentiation.”

“I think there's a massive intellectual failure when you equate using more with value, right? Because that's just not how the economy or business works, right? Like, I think that, and the reality is, I think the iron, the likelihood, Britt, is you will use more…”

“I do think the LLMs are going to collapse into their own weight if they can't figure out how to pay people and deal them into the economy.”

“Publishers have to build it themselves, and in no way, shape, or form should be turning that data over.”

“The funny thing about LLMs is, like, LLMs are excellent at this. Like, if you just, like, download your Twitter data right now, and you're like, hey, can you just, like, give this me in, like, one flat file? Like, they'll never give it to you that way. They'll…”

“I think we have a lot more research to do to find a breakthrough of similar scale to this LLM one. It's not discounting how big of a breakthrough it is. It's just that the ecosystem has gone to like the max, I mean, to like LLM equals internet in its impact. R…”

“I don't think anyone currently believes there's any sort of like long-term defensible important thing specifically around LLMs.”

“Today's like frontier models are like LLM pluses, right? They're LLMs and then they've got different capabilities kind of glommed onto them, but they're still fundamentally Text tradition kind of things”

“Like when you back it all out, you're like, none of these are actually good businesses. They're kind of like product and growth shims towards something, but no one can really tell you what that something exactly is, right? Or how it's going to work as a busine…”

“Like, I don't think anyone thinks that just expanding LLMs and the current technology was just like, there was an era of people were just like, just throw two orders of magnitude, more compute at it, and the game is over. And like, I just, I think that narrati…”

“I only, like, honestly, I threaten them because they do better work with threats.”

“LLMs are only going to carry us so far into AI, but all of these labs are not going to find product market fit and business value. Like it's just not going to happen.”