Note: Episodes listed below are ordered based on how likely they are to match your search request.
"All the kind of things you know and are coming to love about llms, they're able to do in a really nice intuitive way. I mean, the power of llms is not just in the expressive capability of the raw model. It's also how you integrate that into a sort of user interface, into a tool that you can use for solving particular tasks that you do. So for example, if you have a bunch of stuff on a team, you have a bunch of documents like wikis and projects. The LLM knows about all of that."
"Until recently, some of its biggest users were academics exploring topics like online hate speech. And then generative AI came along. I've been told by researchers that llms would not exist if it were not for common crawl. Llms stand for large language models, the algorithms behind AI. Like Chat GPT, llms need to ingest huge chunks of text to learn the rhythm and structure of language so they can write your term paper or wedding vows for you."
"But I understand that there are some private data. There will necessarily be something to do with, with this kind of use case. But will it become, like, super easy to quickly fine tune a model? What are your thoughts on the future of llms? Is it, like, basically where I'm trying to go is, do you think llms will become powerful and easy enough to use, that you will only require them, or is it still, like, just a small part of a bigger product that will all make it work much better?"
"Do you think about llms generating mojo code and helping sort of. When you design a new programming language, it almost seems like, man, it would be nice to sort of almost as a way to learn how I'm supposed to use this thing for them to be trained on some of the mojo code. So I do lead an AI company, so maybe there will be a mojo LLM at some point. But if your question is like, how do we make a language to be suitable for llms? Yeah, I think the cool thing about llms is you don't have to."
"So actually the next frontier for llms is in enterprises. And I believe the best llms for this next wave essentially will be enterprise llms. And the enterprises that I'm seeing that are at the forefront of this, they have started to adopt techniques around fine tuning their models to personalize models for every single user and every single customer of theirs, where the data doesn't actually leak between those customers, but the model is highly personalized according to their access controls for data and just any kind of complexity around their data slices and use cases. So that's what I'm seeing. And it's honestly really exciting to see this because this will further democratize all the possible use cases that can be out there far beyond what I can, I know, even imagine."
"Still to me, an open question. What do you are the interesting limits of llms? You know, I don't see any limits to their form. Their form is impressive. Yeah, yeah, yeah, it's pretty much, I mean, it's close to, well, you said."
"I said, I asked Chat, GPT, give me an explanation of how llms work in no more than three paragraphs and keep it simple. And it explained itself pretty good, I thought. Mm hmm. Yeah. It's really, that definition actually is very good that it gives, again, taking in information from books, articles, websites, and then training the computer on that knowledge and then remixing that content."
"And so I think planning the ability of llms is a big one, and that'll get better over time. The last one is maybe a little bit more vague, but I think even just as builders, we're still figuring out the right ways how to make all these things work. What's the right information flow between all the different nodes in order to get those nodes, which are typically an LLM call to work? Do you want to do few shot prompting? Do you want to fine tune models?"
"These LLMs, now, they have something called function calling capabilities. And so when you're seeing a lot of the more advanced stuff coming out of builders, what they're doing is they're using the LLM to perform reasoning. And then basically, in that reasoning task, they basically allow it to have access to a selection of tools. And the LLM decides. So all of this is usually happening in the background."
"And llms are a domain that needs full stack innovation and computer to run on. And this is a data center scale distributed computing platform. Data center is becoming the computer and apps run on the data center computers while being consumed at the edge on personal devices and different kind of form factors. So we have a multilayered stack of offerings and partner offerings that developers and enterprises can engage with us on. At the very bottom, we have the compute infrastructure layer, where we continue to innovate on with specifically things like hopper architecture."
"The software, right. It's running this os in the cloud that appears to be llm style that can go take actions with other llms. Yeah, I think that's. What about just the nuts and bolts of it? Like how do you put contacts into it?"
"Now, one thing that's not clear is. Exactly what the output of this will be from CNBC. Quote OpenAI and Common sense didn't say. How llms will be tweaked to help aid educators or teens. Altman said llms customized for educational purposes."
"As you guys see in the tweet that I just posted in the chat, made a point today that LLMs are emerging not just as a chat bot, but as a kernel process, meaning a new type of operating system that can do input and output across different modalities, can interpret code, can access the Internet and information, and then can render things in a visual way or in an audio way that the user wants to consume it. So as a result, LLms become the core driver to a new type of computing interface. There was a paper published, and I'll share the link to this paper here as well, and we can put it in the notes. It's not worth pulling up on the screen, that showed that using llms in autonomous driving can actually significantly improve the. Performance of the neural nets that the."
"The LLM itself can't do it, but it can write a program. Got it. In some ways, it's a part of a trick. It's still useful. That's what data scientists do with data they write programs to analyze, especially when it's at scale."
"Llms, if you domain adapt them, right? That is to say, you train them on your data and have them learn your specific behaviors that you want to see correctly, can be very good at also taking any slower context. I look through this query, I have this product, are they relevant together or I have this product, what are the relevant features? What are the attributes? What are the sorts of structured features?"
"Basically a list of tokens that are really similar to the kind of tokens that typical LLM takes as an input. And then you just feed that to the llama in addition to the text. And you just expect LLM to kind of during training, to kind of be able to use those representations to help make decisions. I mean, there's been work along those lines for quite a long time, and now you see those systems, right? I mean, there are llms that have some vision extension, but they're basically hacks in the sense that those things are not like trained end to end to really understand the world."
"And cool thing is, I think llms are a piece of it, but there's all these other things that need to be built around it. I think Llms are there to do a bunch of really great things, but all these other pieces around, we have the smartest entrepreneurs in the world working on things like that. So that is what gives me great pleasure today, is meeting people that are building these types of experiences. It sounds like you think our attention is maybe missing key spots around this session, rightfully so. With these incredible foundational models."
"And it's also one of the first cases that actually works really well. It's the most again, I really love this use case, but it's the most trivial one to apply with this new LLM technology. LLM were trained to complete the next lines of codes according to a context, and this is exactly what code completion. And event generation does. They're less trained and well designed per se, purely by themselves to deal with does this code actually work?"
"I feel like I may be underestimating this, but I'm just going to put it out there that I think it has some potential. One of the evidence points that it doesn't actually matter that much is that perplexity has at online llms for three months now and it doesn't perform great on LM sys. It's like number 30 or something. So it's like, okay, it helps, but it doesn't give you a giant, giant boost. Feel like a lot of stuff I do with llms doesn't need to be online."
"Welcome back to the AI Breakdown Brief, all the AI headline news you need in around five minutes. One of the big trends in LLMS right now is to try to put them in new contexts rather than just being a cloud based bot. Specifically, people are looking into things like personalization, how they allow individuals and businesses to have llms that reference their own data. Two, many companies and products are also looking at how to get llms running on mobile phones or pcs in either a fully offline or hybrid cloud on device sort of way. This is, of course where Apple's efforts are."