Guillermo Rauch: recommendation

Transcript context

…That's right. Well, something that I noticed that I loved at the beginning when you were doing the prompting and that prompt improvement feature is it basically is best practices to make it look good and look better. Which I think is one of the more interesting, I don't know, levers to working with AI is it just has best practices to help you build things that are beautiful and also feels like there's this opportunity of just helping you figure out if what you're building is at all a good idea. "What is the problem you're trying to solve?" It feels like there's a PM1 pager step that should exist. Like, "How do you know this is a problem? What have users told you? How many people have told you this?" Things like that. Yeah. There's something to be said about the fact that over time we're more and more peeking into the mind of the AI. That in itself is becoming a killer feature. So the Deepseek stream, the thinking tokens moment was a very big moment for our industry, I think. Because OpenAI did have the technology, but they decided that for competitive reasons, which, it's a reasonable think to think, no pun intended, they were going to withhold it. And also it wasn't clear that there was going to be product end user and product utility. But when Deepseek hit, it was very obvious that people really liked the idea of understanding how the AI thinks and influencing where it should go. We've gotten actually amazing feedback and bug reports where people actually specifically point out, "Look, this is where the AI went wrong. Please fix it." So the more people we get on this product, the more thumbs up, thumbs down, the more user feedback we get. And by the way, I'll tell you for people out there building products, my number one guidance or piece of advice I would give to any startup founder was, "Create a lot of opportunities for people to give you feedback inside the product." I drew inspiration from Stripe. And this was amazing for the early days of Vercel, there was a feedback button with a very slick inline form, with four emojis that would allow you to decide how you were feeling about the feature, the product at that very moment. And that would go straight into Slack. And we were building day in and day out, just streaming users' thoughts right into our consciousness. And maybe we would get, I don't know, tens, hundreds a day, especially the early days, maybe a couple a day and whatever. When you're building AI products, it's a constant stream of user feedback. So for people that are thinking about not building AI products, it's going to be hard to compete with something that has such a tight feedback loop with users. The whole idea is to capture users' feedback so the next iteration of the model, the prompt, the fine-tuning, the examples, the rag is better. And one of the things that Vercel has done as a result of this insight is we've open-sourced a lot of what makes v0 work. So let's say that you wanted to create the v0 for doctors as an example. You can go to vercel.com/templates, and you can clone a ChatGPT template that basically follows all of the best practices in the world for really high-performance, awesome UIs, and now you can go out and build your own AI products. We've also open-sourced the AISDK, which is the foundational plumbing of v0. It allows you to connect any model and generate UI from its responses, not just output text, but actually generate UI. So maybe because I love showing stuff, I'll just really quickly show you this. Okay, cool.…

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