Amjad Masad: preference

Transcript context

…Imagine a place where you can find all your potential customers and get your message in front of them in a cost-efficient way. If you're a B2B business, that place exists, and it's called LinkedIn. LinkedIn Ads allows you to build the right relationships, drive results, and reach your customers in a respectful environment. Two of my portfolio companies, Webflow and Census, are LinkedIn success stories. Census had a 10x increase in pipeline with a LinkedIn startup team. For Webflow, after ramping up on LinkedIn in Q4, they had the highest marketing source revenue quarter to date. With LinkedIn Ads, you'll have direct access to and can build relationships with decision makers including 950 million members, 180 million senior execs and over 10 million C-level executives. You'll be able to drive results with targeting and measurement tools built specifically for B2B. In tech, LinkedIn generated 2 to 5x higher return on ad spend than any other social media platforms. Audiences on LinkedIn have two times the buying power of the average web audience, and you'll work with a partner who respects the B2B world you operate in. Make B2B marketing everything it can be and get $100 credit on your next campaign. Just go to linkedin.com/podlenny to claim your credit. That linkedin.com/podlenny. Terms and conditions apply. Let's go down this thread actually while this is happening, just like what allows for this to be possible technology wise? What is the stack? Whatever you can share that enables this to exist. Yeah, for sure. First of all, it's all the abstractions that we built. So the way a Replit works is the very bottom layer. It's our runtime. So this is the operating system, this is the package manager, this is the language runtimes. We built a system that is able to install packages in any language, including native packages. So the AI, anytime, it needs a package. I can go here and show one of those. By the way, the AI can take screenshots as well so that it checks it works. So here you can see it is taking screenshots to make sure that the homepage is rendering. Here, you can see it wanted a drag and drop library, and so it installed that. And so it has access to all the packages across all languages, including Linux and all of that. And then the layer on top of that is the editor and the infrastructure that runs the editor, including what I described as the multiplayer editor. And then we expose all of that infrastructure to the AI. And there's almost like a new discipline called AI Computer interfaces. So sort of like HCI is now a ACI and turns out LLMs need interfaces that are actually quite different than humans. They're trying to make them use human interfaces like Anthropic's computer use, but those are really expensive and you need to process all this images and video. So instead, for the shell for example, we give it sort of a text representation of what the shell is doing at a certain increments for package installation. We give it a certain tool for editing. We give it an editor tool that when it's writing the code, it's getting feedback on whether there are errors or not, similar to what a human sees, but it's actually old text just to make it easier. So that's AI computer interface, and obviously, all of that is sitting on foundation models. So the improvement in foundation models has allowed us to build this. The most important model that we use is the Sonnet model from Claude, from Anthropic, and it is the best model at coding. So that's the model we use for coding, but we use models from OpenAI as well because a multi-agent system. And so we have models that are critiquing. We have manager editor model, and we have a critique model and different models will have different powers. We also train some of our models, like the embedding model for search is something we trained internally. So I actually wrote about it back in '22. I said it's going to be society of models, like products will be made of a lot of different models, and it's quite a heavy engineering project. To say the least. We were talking offline and you said you've been working on this since 2009 when you first built the first idea of Replit. Is that right?…

Stored transcript either side of the excerpt. The highlighted words are the published quote; the surrounding text is unedited source, never generated.