Large Language Model

“You know, I found two new AI phrases. I don't remember both of them, but one of them is in our increasingly online world, when someone's writing about online culture, or something that happened on the internet, or, like, with the boom of social media, like, th…”

“There is no attorney-client privilege between you and an LLM, and so if later it does become a lawsuit and in Discovery they subpoena your LLM chat, your conversations, they can pull those into Discovery much like they're an email or a private message. And so …”

“What happened was the LLM user showed severe deficits, quote, in memory and essay ownership, with 83% being unable to quote from the essays they had just written.”

“There was an idea of continuous learning from over the last couple of years that said that you would have a, like a literal LLM like model where all of the learning happens internal to this closed loop system. That technology has not been developed. It doesn't…”

“We would literally take a large language model and we would turn it into the robot brain, which this is like very simply stated, but honestly at the time was like a kind of a crazy idea to do because everybody was just trying to make robot brains in many other…”

“I mean, we just got, you know, hammered in the, by like all of the intellectual giants of the field for saying that we were going after AGI. And then when we started really focusing on large language models, we got hammered again saying this is completely ridi…”

“Current LLMs, like you're using them as an ordinary person, for the most part, they are helpful, and they try to solve your task that you assign them, or give an answer that is, sometimes they hallucinate, or maybe deceive a little bit, but broadly speaking, t…”

“And so there was a recent paper by Anthropic looking at the existence of a kind of global workspace inside these large language models. This is the idea of there being a kind of almost like a stage inside a mind where some small subset of all the information t…”

“I think about an agent as an LLM call that is wrapped in machinery that's used to construct context. It's really a big context construction machine. And also it provides a way of executing actions.”

“So one of the mental models we have is I often tell folks like, look, you should treat the underlying, especially if you're building agentic products, you should treat the underlying LLM as like a very athletic middle schooler or high schooler, right? They're …”

“So many tasks instead of just knowledge retrieval require actually firing up a browser, scraping it, writing some code, downloading things, you know, setting up an actual service and workflow as opposed to just something that can be done within the context win…”

“So there's some activation state that by definition is going to be biased to that current trajectory. And so maybe for review, you want an uncorrelated trajectory, right? Where it's like a new box and it's just a smart.”

“Serving large language model is a fundamentally different problem. Because serving it requires to run it on accelerators like GPUs or TPUs, and it is a computationally intensive process that will require a lot of engineering and ensuring that for each request,…”

“You don't want to mess with the model weights because you run a chance of making the model dumber at something else for the purpose of giving it vision.”

“Speech is autoregressive. You effectively I mean, this was even back with, like, the Orpheus architecture a year and a half ago. You just add a bunch of waveforms to the vocabulary so that the LLM can output tokens that represent those waveforms, and then you …”

“Like, a big reason why graph engineering matters is it makes quality less dependent on someone remembering a perfect prompt to ask their LLM. It makes reviews way more consistent. It makes delegation in general way cleaner. It makes approval way more explicit.…”

“One of the early observations that we made was that these models are trained on so much of human behavior data, sentiment data that were expressed on the web. So if you poke at them sort of at the right angle, you could actually extract a lot of realistic huma…”

“I think people, as they become more uncertain with their politics, are going to go more and more and more to large language models and say, who should I vote for?”

“It, it's, it, a large language model and its training and inference phases are completely opaque. You can't say, why did you give me this solution versus this one? And if I just change a variable slightly, what will be the outcome? Non-reproducible results, no…”

“Yeah, so when the phrase, when the question is phrased like this, does an LLM, a large language model by itself, lead to AGI, the answer is a clear no.”

“There's a lot of categories where an AI, an LLM-enabled app is really good, and it's the next generation, and it's obvious what it should be, and customers are going to buy it. But it's not like going to scratch the corporate itch to use, you know, it's not go…”

“The LLM itself Is not a great product for most people. Usage is a mile wide and an inch deep, and you have this kind of polarization between people where this really, really works, and they really, really have product market fit, which is basically software de…”

“I would not be That concern with LLM is replacing every kind of a white color jobs. That's not going to happen. It's not that smart. You keep making silly mistakes. It's not very good. Actually, it's not very good at reasoning. It's very good at collecting dat…”

“It is so cheap to run a single inference on an SLM versus an LLM. So that gap will always be there.”

“I'm skeptical of LLM-mediated permissions for stuff that is At the sandbox level, because you do want hard boundaries. Otherwise, obviously someone can exfiltrate stuff.”

“Large language models, those that are meant for generalized usage, they're usually trained mainly on English, and maybe a couple of languages, and if you really want to go broad, if you really want to go multilingual, you need models, you need the data and the…”

“In a post LLM world, we feel MBAs will definitely have a lot more value especially engineers doing MBA as an MBA is lot more philosophical thinking understanding about business.”

“I think, interestingly, my bet is also on a lot of good arts graduates being a lot more valuable now. You know, now you're coding and LLM, you know, coding is going to be more sort of commoditized.”

“The podcast world is still sort of highly naive to serving content to the algorithm. Like, you know, if you get deep into the, like, Twitter group chats, especially now that they've posted the algorithm, like, there are all these, like, very specific things th…”

“The LPU and GPU, as mentioned, they, they're better at different parts of what's, what's called the decoder layer of an LLM. The GPU is better at the attention portion, and the LPU is better at sort of applying the weights, which is the thing that gets trained…”

“The lag time between I've had amazing invention and revenue in the kind of LL, core LLM space is a year, or two years, or three years. The lag time if you have a medical invention, an idea, and turning that into real meaningful economic value is 10 or 15 years…”

“But now you can send LLMs in because the beauty of code is self descriptive. It goes in, it understands what everything is doing and why, and then it can also very easily convert it. And then the key is like, how do you write harnesses to do validation and rec…”

“Somebody discovered recently that it's great at suggesting off-sheet references. We found some really good off-sheet references by saying this is who we're doing diligence on. Who in my network would you talk to?”

“If we can automate all of these document centric tasks at machine speed, patients get onboarded and assigned a bed in the emergency department much faster. Nurses are freed up to deliver care with more of their time. Referrals can happen at machine speed so pa…”

“I think the only right solution in this case is, Even though you have a local model, or even though you're using the enterprise safe version of the model is just to eliminate the data completely. So if you don't have anything sensitive that goes to AI in the f…”

“Feels like you can fix 90% of a design slop, which is not a capability gap. It's more like a contract gap in what your harness is telling an LLM to do versus what your user is saying.”

“If an LLM already knows about something, it should not end up in your, you know, skill or taste file. That is absolutely useless context, right?”

“There is an LLM that's reasoning and planning at the core. But there is a harness around it that provides a lot of determinism. And so that determinism is adding in governance, security, trust, integrations, permissions, and that's really what's going to make …”

“The durability of these so-called best practices is so short, like the prompting tools and guidelines and context management, like stuff, stuff that we were writing about six months ago, 12 months ago, 18 months ago is now completely wrong.”

“It's crazy, like, you teach this thing to predict the next word, and somehow, if the next word is hard enough, it has to learn to really plan ahead, and it has to learn how to do all of this.”

“The world may not need 10 large language models. Maybe it needs four will be winners, and two will be very big winners, and the other two will get by, and maybe it'll get reduced over time to two.”

“You know, one of the, you know, when you get, when you go into some of these large language models, you don't know the thought process. You lose intuition in these things. It used to be when I started out in the business, You know, people be shrieking each oth…”

“Because it's been going around so long, it's kind of in the latent space of these models already, is if you say, you don't need to invent this whole new set of terms, you can just bolt on this framework, this system that's very, very flexible, very composable,…”

“You build a LLM completely from scratch. You use PyTorch, but you build one from scratch that, you know, can be a chatbot. You train it on data. You RL it to solve math problems with tool calls. You do all of this. And this is a undergrad level course.”

“You have this code this code to build a complete large language model that can train on a large data set and learn to speak, runs on GPUs yes, eventually is trained with RL and tool calls. That entire set of code, probably two to 300 lines of Python code.”

“And I've seen draft legislation that imposes like strict criminal liability if the LLM Is even capable of generating certain kinds of content.”

“Over, I think it was about a year and a half ago, over 80% of our code was LLM written, and now it's much higher than that”

“You can't just go defer to the large language model and say, hey, based on what you know about this user and the data I have available, what's the next ad to see? That wouldn't work as well as a custom model built for this purpose.”

“Now, if you want to do these predictions, quite frankly, then the challenge is large language models are not made for this, right? The way how they, you know, generate just one token after another, essentially, in a sequence-to-sequence modeling. I mean, They'…”

“And that's essentially what a large language model is. It has no regard for the truth. It wants to please you. If what pleases you happens to be true, great. But if it's not true, then it doesn't care one way or another.”

“The YouTube Rexxus is LLM-based, and they- Is it really? They obviously, yeah. That's cool. It, they re, they tokenize every video, and put it in a code book, and then they train a LLM on it, and then feed in your context, just like a regular LLM, and ask it t…”

“So I find that that's helpful just to challenge any LLM giving it references.”

“So this is probably the best deck I have ever seen created by any LLM period. Period.”

“The homeschool one was the most obvious one to me, because I was like, I want you to literally, like, use this curriculum or use this book as a reference point, but then I noticed how well it worked, and I was like, okay, what if I take this agent over here th…”

“I believe open source is going to win the day on the large language models and take 90% of the token usage, and I think the entire frontier model space could be undercut by open source”

“We think about them almost as like an orchestration layer. So that's sort of a co-pilot assistant, but also like a system that can direct experiments. And it's almost, it's orchestrating other specialized models as well. So we do construct neural nets that are…”

“There appears to be a very strong correlation between how good their drawing of a pelican riding a bicycle is and how good they are at everything else. And nobody can explain to me why that is. But as I started looking at these things, I realized, wow, The bet…”

“I think the UTM chat GPT is around eight percent on the traffic, but when you talk, I think it's more around 20% traffic coming from LLM today.”

“The highest bandwidth API for an LLM is just chatting to it, is just saying like, I have Gmail and I have these folders that are really disorganized. And what I really want is to be able to come into my inbox every day and really know what's important, get rid…”

“And it's hard to just scrape it. And it's one of the things I love the most about it. It's hard to just, you know, scrape all that information, say job is finished, and then put it through some LLM or something. Well, A, that data is always changing. B, it's n…”