If you're interested in a high level overview, check out our recent blog post. Otherwise continue reading for a technical overview of LSD.
If Google provided a map for the internet highway, we're building the GPS. Now, what do we mean by that? To help understand, let's consider the problem that Larry and Sergey solved with PageRank - there's a ton of stuff on the web that people want to sift through. Prior attempts at grokking the internet led to keyword search which, admittedly, works in an ideal world where people don't game websites to show up higher in search results. But it's not as though there's an entire industry devoted to that, right?
In an ideal world, thanks to the wonders of technology, there are robots working for us so we can focus on pursuits that accentuate the difference between man and machine. Instead, we live in a world with:
What LSD provides is an "operating system" for the web that seamlessly ties together how you interface with the internet into an intelligent agent; one that can both answer questions and execute actions reliably instead of gambling with a generative model's output. But how can a tool that's used for scraping hacker news be helpful for creating an "intelligent" agent?
Such as folks who get hired to spend hours verifying Excel arithmetic by hand, why we still have the most menial jobs that should have been automated decades ago is simple: blame attribution - who or what is "responsible" for seeing an outcome get accomplished. As much as Silicon Valley has gotten people excited about AGI or other romantic narratives, we haven't heard of a company do a mass layoff because knowledge workers are now rendered obsolete. While you can replace employees with a chatbot prompt, you wouldn't have a way to guarantee each task was appropriately handled or managed when it couldn't be handled.
For those who want to coerce a swarm of generative agents, there are developers working on solutions like LangGraph or CrewAI. However, how about those who want to be able to intelligently teach agents and correct them as the world evolves? In our stack, we provide a solution by leveraging a human in the loop who can easily annotate the details agents and developers share via our browser's click UI.
By threading together cohesive workflows and the framework for an ecosystem of actions, you can finally stop thinking about websites like they're paintings in a museum where you can only look but not touch. Instead, think about how you want to remix the web to define your own internet. If you've made it this far, you might as well get started!