Shawn Wang: evaluation

Transcript context

…No, it is a much bigger rewrite. I'll give you a concrete example. Suppose you want to have an agent that observes your Zendesk tickets, and it's like, hey, every time you receive a Zendesk ticket, I want you to check my knowledge base, so it's like a RAG module and whatnot, and then answer the ticket. The way it used to work with Lindy before was, you would type the prompt asking it to do that. You check my knowledge base, and so on and so forth. The problem with doing that is that it can always go wrong. You're praying the LLM gods that they will actually invoke your knowledge base, but I don't want to ask it. I want it to always, 100% of the time, consult the knowledge base after it receives a Zendesk ticket. And so with Lindy, you can actually have the trigger, which is Zendesk ticket received, have the knowledge base consult, which is always there, and then have the agent. So you can really set up your agent any way you want like that. This is something I think about for AI engineering as well, which is the big labs want you to hand over everything in the prompts, and only code of English, and then the smaller brains, the GPU pours, always want to write more code to make things more deterministic and reliable and controllable. One way I put it is put Shoggoth in a box and make it a very small, the minimal viable box. Everything else should be traditional, if this, then that software. I love that characterization, put the Shoggoth in the box. Yeah, we talk about using as much AI as necessary and as little as possible.…

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