LLMs

“The only thing that mattered for the last two years is LLMs do code, and code is a half a trillion dollar industry.”

“Anything that can be reduced to code will be done by it.”

“I do still believe that, like, AI agents are not that good today actually at doing something that is very relationship driven. But they're getting much better at it, right? So, like, the problem is when you look at LLMs, they look at the distribution site, and…”

“If it can solve arc AGI and it's not necessarily crushing on these economic factors and these just kind of general rote work things that we would like it to do it shows that instead of being general, it's very spiky intelligence and hence much less useful.”

“As opposed to having humans in your compliance and security team over time, label data and decide what goes where and what can be accessed. I think we'll just start to trust LLMs to do that.”

“I think the LLMs will make many fewer of these mistakes than humans pretty quickly.”

“One way to think about LLMs, right, is they're just, they turn energy into, like, GDP, right, or into revenue.”

“So instead, what we're using is we're using like all the LLMs to help come up with the plans, but the plans themselves Of how you're going to do a recovery in a cyber scenario are deterministic and then are explainable to the customer.”

“In today's world, the words that you say and how you portray something, that becomes a part of the story that the intelligence systems that we're building ingest. And they, they're like world model, LLMs, and the tools that we're going to use to do work on a d…”

“So today already there are various reports, including Gartner, that says that 50% of consumers have moved to LLMs to make buying decisions. 90% of enterprises have moved to LLMs, especially in high research categories like education, travel, healthcare, softwa…”

“Every single brand, every single company, every single SMA is spending a combined trillions of dollars on advertising every year, right? A lot of that is going to move to LLMs is our thesis.”

“I don't think that will succeed. It's a bad approach from a technical standpoint, because I think these labs and LLMs will figure it out pretty soon.”

“I think one of the things that's happened with LLMs, we're now very firmly in this world of, hey, it's all about building the software that goes around the core LLM, the harness as we speak, as we talk about it.”

“I feel like I was a little too late in the two point of transformation. Maybe we should have done a year earlier. So I could have done this sooner. I think I, Underestimated the power of LLMs and the power of where, where the very eyes going in general.”

“We run into it a lot at the application layer where people build wrappers, so much, so to speak on top of these LLMs, like the antithesis of building a moat, right? You need some underlying IP and tech, your own model, your own data set, something else that is…”

“And, you know, I think this class of models is going to be like just as big, just as impactful as LLMs, but it's almost like the compute market, like kind of doesn't realize that yet, both in the capacity sense, but also in like the software stack sense.”

“Repetitive tasks in controlled environments is, like, more or less solved, and, like, you know, that'll continue to happen. But having tweaked tasks in real-world environments, that's where you need the AI. Like, that's where you need the AI plus the robots. A…”

“So as an example, we've built our own inference engine. It's pretty important because we think that these interactive real-time models are going to need to run in a pretty different way. So having the ability to have your own engine that is not designed for LL…”

“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.”

“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…”

“So where I think you can win and where he's winning right now is that he's sort of the one of the first in this space. And then his product will be recommended through LLMs for agents to use. And then that could be where the business goes to a million dollars …”

“Now with LLMs you're able to reverse engineer that patch in a fraction of the time. So that part of AI and cyber is a legitimate step up in capability, which is the attacker's ability to quickly revert reverse engineer and weaponize a flaw in a patch, and then…”

“Reddit is being eroded right now by people manipulating it for like LLMs to rank on them. but this is a way like you buy people like me that are basically creating the dead internet. Let's be real. It's already happening.”

“The kind of insane thing about LLMs is that agility used to mean that you would do all those things over, let's say, a six to 12 month cycle, and now it's become possible, keyword being possible, to do them over a day to week cycle instead.”

“The main thing here is that LLMs are really amazing at code, and the amount of economic value that can be created from that dwells to everything else, right?”

“What these LLMs actually do really well with is taking a massive stream of consciousness where you just keep talking and talking and talking, so I'll give it a one to two minute prompt, then I let go, and it has changed everything.”

“These LLMs are obviously amazing and, but they do lack like the vertical specific context and stuff like that. So you're seeing MCPs, you know, sort of, you have the mob in MCP for designs. You're seeing the idea browser for business building MCP. You're seein…”

“Yeah, probably LLMs are going to run out at some point and they're not AGI and okay, we have maybe another 30 years of AI winter or something and then like the next paradigm really is actually the thing.”

“The demand that we see for something like that is actually not for LLMs. Although sometimes people want to run evals and do synthetic data prep and there it makes sense. But it's from a lot of non LLM companies like people who are doing computational bio, like…”

“Or you can even train a model which is able to natively understand crystal structure representation, and like image modality is a modality for LLMs these days. Crystal structure can be a modality for LLMs as well.”

“If our American LLMs know nothing about security, they will create more security flaws. So we cannot create a standard where American LLMs are dumb about security. That would be a humongous own goal.”

“We cannot set the standard that US AI models can't find bugs. That is a terrible, terrible, terrible standard. If you have a, if you are writing software with an LLM, it has to be able to understand what a bug looks like so it does not write those bugs.”

“And a new way to do it is actually have the LLMs just write the code you need right now. And then you write 10 or 20 times less code, but it's way more customizable. It does, you know, more and is more awesome. Between token maxing and even like, you know, don…”

“It's a moat destroyer when LLMs will lift you from one vendor to the other, and I, you know, I don't even, like, I literally was doing a pitch this week, and the founder was going on and on about their moats, and I immediately didn't want to invest. Like, I ju…”

“This belief that the LLMs are gonna eat up every possible business, I just, I don't think is true. It's, I mean, cursor flies in the face of that.”

“I think the analytic world is getting reshaped already where you can see people like Snowflake or Glean or Databricks. All these people boast enterprise data lakes where you can bring the data and run LLMs against it and get you much more synthesized analytics…”

“Yeah, probably. I mean, I don't know about Anthropic, but on LLMs.”

“And I think the biggest pitfall of LMs is you have no sense of How much training data the model has seen for the exact thing that you're asking it.”

“I think we've already crossed a threshold. Yeah. For which the context links Is larger than any context that human can have. Right. And I think that's huge because, I mean, just think back two years ago we had 16 K context link. Right now we have a one million…”

“LLMs are going to be very good at anything where you can describe how people do it. And where what you want is the way anybody would do that, and not so good at where you can't really explain why you did it like that, and where you're doing it differently to t…”

“I like to say, like, all the cool stuff that we're doing on the, with LLMs now, is really based on just the text of the internet being absorbed into these models, which is incredibly powerful already, but they don't know shit about the real world. I call them …”

“What we saw is the moment you send the result with the repair logic, right after that, the third tool call is fixed. Instead of, you know, it all of a sudden becomes super smart. It understands like, okay, I got the result, what I was looking for, and I'm gonn…”

“Like if they are seeing a lot less tool call errors, they're much more creative. They are, they can explore a lot and they can continue a lot longer.”

“I personally don't use OKLCH, but apparently LLMs are really good at it. And if you see them using HSL or something, they are, they don't actually are able to control the lightness in HSL very quickly, but on to human eye, it's very, very easy to see like this…”

“They're totally putting our business on steroids. Like, without LLMs, nobody would be talking about our ontology, about Apollo managing secure exploits, about our ability to manage an enterprise, essentially, as turning all these companies into FDEs.”

“The thing LLMs have been somewhat unproven as of yet is the ability to actually trade stocks. You basically be a pod manager in a hedge fund world and outperform humans. The record on that isn't there yet.”

“I find AI in like what most of this like launch here is, is actually like less building forward. It's actually like. A recursive loop backwards. I'm always looking at what had happened first, like go back through the week and tell me what we did, what worked, …”

“A year ago people talked about the commoditization of the LLMs. And frankly, it's gone the opposite because as you start building an agentic layer, and we've all started to use this word harness, but the harness is what brings the context, the memory, right?”

“I feel this is actually, this part of long contacts is a little bit ahead of the LLM part.”

“And what's cool about LLMs I think is not that they're artifact generating machines. They're okay at that, but like take them out of their distribution and all of a sudden you're in big trouble, but they're incredibly good teaching machines. There may be the b…”

“And that has to do with the mechanics of the attention mechanism. As the contracts gets larger, you're literally spreading the attention over more and more and more data. But also, it's learning from all the bad examples it gave. So it's actually much more lik…”

“The fact that Grokipedia is a great intervention because Wikipedia is going wonky and their efforts to influence it, and that therefore influences the LLMs because they're trained on Wikipedia.”

“And LLMs are very good at helping you do DOK one, two, and three. And reasoning around that. And then the part that the humans still do really well is DOK four creating new knowledge.”

“The number one biggest finding, I think, for a use of LLMs is that LLMs cause the Dunning-Kruger effect to be magnified. They convince you that you're more capable than you are at while they degrade your actual capabilities.”

“And of course, you know, a use case that probably 60% of the country is engaged in now is to use these LLMs for their own health diagnosis and research and people getting great value out of that.”

“These LLMs and even video models, they don't know how to do that. They don't have the ability to understand friction or weight or contact pressure friction, surface texture. Like they're just not able to do these things. And this is the core of what we need. A…”

“These AI systems need to log the entire trace of how the AI reasoned on the initial input. What were the tool calls it made? How did it interact with the LLMs? How did it sort of ultimately generate the response? And did that response actually meet the user in…”

“Every 18 months, you know, the raw performance of the chip gets, you know, roughly three X faster. Then on top of that, other optimizations, how they run LLMs, how they do all the quantization, all the other clever stuff that, you know, you can read about and …”

“I would have never had the confidence to push on the doctor so hard to get the scan if I didn't have LLMs open during the conversation with the doctor. I definitely would not have found another machine, got my dad scanned, and then figured out.”

“A prerequisite is we have like this entrance exam. There's this gigantic code repo that you need to read very fast. You can't read it with your eyes. You have to use AI. You have to use LLMs to ship all the features, and then you get a test at the end, which i…”