Unaligned #8: Demo of Mindstudio, the fastest way to build AI-powered apps

So making AIs is really simple. You just log in, say new AI, and then you get this AI builder. It's really broken up into three areas. Left hand side, middle and right hand side. On the right hand side there's this live terminal that you can interact with as you build your AI. Here at the bottom we have something called a quick start panel. And so we can start here, let's go ahead and build a blog post generator where we're going to ask the user to paste the URL of something out on the web and then we'll go grab that and we'll convert that into bullet points and we'll send it to an AI to turn that into a blog post. So let's try that. We're going to say generate prompt. And here we can have the AI generate a prompt for us to use. And so I can tell it what I want my AI to do. And here I'm just going to say create blog post because based on bullet points provided by user, I'm going to say generate and this thing is going to go out and sort of write our prompt for us. So we don't have to do that. We could have done that manually. We can edit this after it writes it, but generally it will include a bunch of things that we would have written by hand. And this is now trained on many AIs that have been created with MindStudio and sort of best practices of writing these prompts. Okay, great. So we've got that done. Now we can go to this automation section and create our workflow. And so we'll start first with creating a new node here and say collect input and we're going to ask the user to paste the URL. So we go over here and say plus and say create new. You can create different types of prompts for the user to engage with. We're going to say scrape URL and we're going to call this URL Content. That's going to come back to us. And here we're going to say enter a URL to create blog post about. Okay, something like that. And so now we can go back to our main flow here. And now we have this thing over here. Now we're going to take the output of that, which again we called URL Content. We're going to create another node and here we're going to use the send message block and we're going to send that to a model. And here we can choose what model we're going to send it to. I'm going to choose Claude 2.1, because it's great that taking a bunch of text that's going to come back from that website and making sense of it. And I will say summarize this into bullet points. Only return bullet points. Okay. And here I can give it what to summarize. And we put that into double curly braces and we called that URL content. Right. Okay. And now we're going to take that and instead of displaying that to the user, we're going to assign it to a variable and we're going to call this variable bullet points. And then we're done here. Then we'll create another node here and we're going to do another send message block. And here we're going to use GPT4 Turbo because it's better at writing blog posts, in my opinion, than Claude. And here we're going to say create a blog post based on these bullet points. In here I can simply return back this variable that we brought back here, which is called bullet points. And here in curly braces I can write bullet points. Okay. And so we just created this thing. Now we can preview it by going here and saying preview draft. We didn't give it a name yet, so that's why it's called Untitled. But when we launch it, we get this. I'm just going to give it this Wikipedia entry for Transformers. And as you can see, it's got a blog post that it's creating and we did all of that in a couple of minutes. Now this is obviously a simple example, but you get the point of these automations, these multi step workflows that you can create that leverage different models, the right model for the right part of the job. And some of these models are good at one thing, others at another thing. And as obviously as all of that evolves, this is going to become more and more important to be able to do so again. That's a really simple demo of it. If we wanted to make it more fancy, we could go back here and we could go here to this onboarding section and we could create another prompt, for example, that when the user first shows up, we could ask them, for example, give them long text or upload file. It's up to us. I can just say long text and call this writing style and ask them to, you know, paste in some samples, some samples of your writing style. Okay? And they will paste in these samples. We now have this variable called writing style. And then we could go over here to this area here, create a blog post based on these bullet points. And then after that we can say, maybe these backticks work well, just to separate things. Write it in this style. I need to learn to type. And then we can paste in our variable here for writing style. You get the point. That's how you can take all of these things and collect variables during onboarding and then during runtime, route them all through this thing and end up in a document. And then you can publish it. When you publish it, you can specify your name, description, various icons. You can specify whether it's free to use or you want to charge users for it. You can get the embedding code. This requires some payment plans that we have. Okay, well, I hope you guys enjoyed this demo. And yeah, back to you, Robert. And it's going to go and grab that URL, bring back the content, take it, pass it to Claude 2.1 to be able to summarize it into a bunch of bullet points, and then take those bullet points and pass them to GPT4 Turbo in order for it to take those bullet points and turn them into an awesome blog post for us. And I'm going to paste that in there.