image: thumbnail.jpg
🐙 Lunch & Learn: Let’s Build An AI Assistant With GPT-4o (w/ Joe & Winston of @Posit)
@Posit Channel - https://www.youtube.com/@UC3xfbCMLCw1Hh4dWop3XtHg
🤖 Sign up here to get accompanying workbooks, summaries, and notifs for future lunch & learns: https://www.lonelyoctopus.com/email-signup
✉️ NEWSLETTER: https://tinahuang.substack.com/
It's about learning, coding, and generally how to get your sh*t together c:
🐙 Lonely Octopus: https://www.lonelyoctopus.com/
Check it out if you're interested in learning AI & data skill, then applying them to real company projects!
🔗Affiliates
========================
My SQL for data science interviews course (10 full interviews):
https://365datascience.com/learn-sql-for-data-science-interviews/
https://365datascience.pxf.io/WD0za3 (link for 57% discount for their complete data science training)
Check out StrataScratch for data science interview prep:
https://stratascratch.com/?via=tina
🎥 My filming setup
========================
📷 camera: https://amzn.to/3LHbi7N
🎤 mic: https://amzn.to/3LqoFJb
🔭 tripod: https://amzn.to/3DkjGHe
💡 lights: https://amzn.to/3LmOhqk
📲Socials
========================
instagram: https://www.instagram.com/hellotinah/
linkedin: https://www.linkedin.com/in/tinaw-h/
discord: https://discord.gg/5mMAtprshX
🤯Study with Tina
========================
Study with Tina channel:
https://www.youtube.com/channel/UCI8JpGrDmtggrryhml8kFGw
How to make a studying scoreboard:
https://www.youtube.com/watch?v=KAVw910mIrI
Scoreboard website:
scoreboardswithtina.com
livestreaming google calendar:
https://bit.ly/3wvPzHB
🎥Other videos you might be interested in
========================
How I consistently study with a full time job:
https://www.youtube.com/watch?v=INymz5VwLmk
How I would learn to code (if I could start over):
https://www.youtube.com/watch?v=MHPGeQD8TvI&t=84s
🐈⬛🐈⬛About me
========================
Hi, my name is Tina and I'm an ex-Meta data scientist turned internet person!
📧Contact
========================
youtube: youtube comments are by far the best way to get a response from me!
linkedin: https://www.linkedin.com/in/tinaw-h/
email for business inquiries only: hellotinah@gmail.com
========================
Some links are affiliate links and I may receive a small portion of sales price at no cost to you. I really appreciate your support in helping improve this channel! :)
Transcript#
This transcript was generated automatically and may contain errors.
Hello friends! How's it going?
Spock's daughter? Well, hello! Steve Jobless. That's actually pretty funny. Steve Jobless. Hi! Hello everybody!
Some audio issues. We're good. Ask AI to fix audio issues. Are you using VPN? No, I'm not using VPN. I'm in Hong Kong right now, so you don't need to use a VPN.
Alright, let's just jump into it because there's a lot to go through. I'm so excited to go through this. As you can see, there are two wonderful guests that we have today. I'm going to let them introduce themselves, but very quickly, the structure of today is I'm going to go show you guys the demos that I'm talking about.
GPT-4o got released, I think it was a week ago, two weeks ago at this point. They were demonstrating how cool the multi-modalities are. They have video functions, really natural voice, and things like that. But the problem is that you can't directly access it right now. Most people don't have access to it. Also, if you go through their API, you can't directly call it. That's why we need to build this application so that we're able to have a similar user experience.
That's what we're going to be doing today. After we build that, show the demos. After we do that, then we're going to actually use it and try to replicate their demos to see if they are, in fact, being very honest with us. I bet it can actually reproduce that, because as we know, sometimes tech companies like to make their things seem really cool, but it might not be as cool. Word could be really cool, so we shall see. Any questions or thoughts? Is that the Her one? Yes, it is.
It's the one that they, it was, yeah, at least from the demos, it was honestly quite impressive.
In that case, I will hand it over to Joe and Winston to introduce themselves. As always, guys, drop everything in the chat that you see fit. Any questions, I'll be monitoring the chat as well. While Joe and Winston are going to introduce themselves, I'm also going to drop the GitHub link, so if you want to follow along the code today.
Introductions
Hey, everyone. Thanks, Tina, for having us. My name is Joe Cheng. I am the CTO of a company called Posit. We used to be called RStudio, rebranded to Posit a couple of years ago. I've been with the company almost 15 years now. I was the first employee after the founder. About 12 years ago, I was the original creator of Shiny, which was a framework for R and is now a framework for R and Python that we're going to talk a lot about today.
All right. Thanks. Yeah, I am Winston Cheng. I have been at RStudio and now Posit for 12 years, so not quite as long as Joe. I've been working on Shiny for, I think, what, like about 11 years of that, Joe? I mean, not just Shiny, but that's been my primary thing for quite a while now.
You know, I wanted to mention that Shiny has been around for R for a long time, and we've recently brought it to Python as well. It's not like it's like a wrapper for R Shiny. It's just a complete new implementation in Python. But that's what we're going to talk about today.
Overview of the GPT-4o demos
We can hear you. Okay, great. All right. 116 people watching now, by the way.
Okay, so, yeah, if you guys haven't seen it, like, this is where they introduced GPT-4o. I guess it was streamed 12 days ago. And then the demo parts of it. So, there's like a few different things that they showed.
The first one is around this area. And I want you to tell them a bedtime story about robots and love. Oh, a bedtime story about robots and love? I got you covered. Gather round, parrots. Once upon a time, in a world not too different from ours, there was a robot named Pipe. Pipe was a curious robot, always exploring new story questions. I want a little bit more emotion in this, a little bit more drama. Got it. Let's add some drama. Once upon a time, in a world not too different from ours, there was a robot named Pipe. I really want maximal emotion, like maximal expressiveness, much more than you were doing before. Understood. Let's amplify the drama. Once upon a time, in a world not too different from ours.
So, in this case, they were just trying to showcase the fact that it's a very natural voice.
And then the next one is... Hello there. How's it going? It's going really well. Today, I'd really like your help solving a math problem. I'm all ears. What math problem can I help you tackle today? So, I'm going to write down a linear equation on a sheet of paper, and I'll show you. And then I'd love your help working me through it. But importantly, don't give me the solution, just help give me hints along the way. Got it.
Okay, I see it. No, I didn't show you yet. Just give me help along the way. One second. Whoops. I got too excited. I'm ready when you are.
So, ChatGPT, what equation did I write down there? Ah, I see it now. You wrote down 3x plus 1 equals 4.
Okay, so that was the other one. It's able to go and see it and then be able to help you solve these math equations. So, they're showing the vision capabilities. See anything on the screen. So, I'm going to highlight the code, Command-C it, and then that will send it to ChatGPT. And then I'm going to talk about the code to ChatGPT.
Hey, ChatGPT. Hey there. How's it going? Yeah, it's going really well. I was wondering if you could help me with a coding problem today. Of course. I'd love to help you out. What's the coding problem you're dealing with? Okay, I'm going to share with you some code. One second. Sure thing. Take your time. Okay, so I just shared some code with you. Could you give me a really brief one-sentence description of what's going on in the code? This code fetches daily weather data for a specific location and time period, smooths the temperature data.
So then they just generated this plot and then asked you to describe it.
Mike, she wonders if whales could talk, what would they tell us? So this one is going to show the language capabilities using different languages.
It looks like you're feeling pretty happy and cheerful with a big smile and maybe even a touch of excitement. Whatever's going on, it seems like. Okay, yeah, so those are the ones that they demonstrated. And what we're going to build right now is how we're going to try to replicate those things.
Planning the architecture
All right, any questions? I will only be impressed when it sounds like Matt Berry. I don't know if that is going to happen anytime soon. So I'm sorry that you're going to be disappointed.
Yeah, absolutely. Sorry, I was just going to say, let's just get into it. And feel free to ask any questions. The GitHub is linked. It's pinned on the top of the chat right now. And also, like, real quick as well, if you would like to, I will also email everybody these. And then also there's templates that Joe is going to provide. So you can also make this app and some other apps as well. It's like a really quick plug and play. So just sign up to this email mailing list. And then this is when you get weekly emails where basically I send the resources from the livestream and then tell you about the next livestream. That's it. It's completely free. Don't worry.
And like Tina said, while they demoed that, the APIs are not actually available just yet. They have made GPT-4 available, but only via the same chat completion model that we've had for a while now with GPT-4 and its predecessors.
So the question that we had last week when the three of us met to talk about this is, could we do our own homebrew version of this? Could we take the pieces that actually are available to us now and create an experience similar to this AI assistant? So this is what we came up with in that call last week. On Friday, I think it was, or Thursday of last week.
So we start out with some kind of video input, right? We have a person talking into the camera, and we run that through a program called FFmpeg. And that will split up the movie into a speech track, just the audio, and a bunch of JPEGs for the video. That's necessary because the only thing that today's GPT API understands is textual prompts and images. Those are the only kinds of inputs that you can give it. And the only kinds of outputs it can give you are text. Well, it can also make function calls, but mostly text.
So this speech track is not quite enough for us to feed into GPT. We have to boil it further down using the Whisper API from OpenAI into text. So with a handful of JPEGs and some text, now we have input that GPT-4 can understand. And so we feed that into the new GPT-4o. We get textual responses out, and we'll feed that into one more OpenAI API for text-to-speech. And that results in audio output. So that completes the loop. We say something into a video camera, and we get an audio response back.
So this speech track is not quite enough for us to feed into GPT. We have to boil it further down using the Whisper API from OpenAI into text. So with a handful of JPEGs and some text, now we have input that GPT-4 can understand.
Did I miss anything? Does that make sense? I think it might be worth pointing out that I know you mentioned this earlier, but just in case everyone out there didn't quite catch this, the OpenAI app can do more stuff than this right now because they have access to more functionality than is exposed via their public API. And we can just use the public API. So this is why we're doing it this way.
Other Joe says, I like your diagram. They're actually teaching us how to make Skynet. Exactly. Precisely.
Okay. And by the way, one thing that you brought up, Tina, last week is, could we take all the OpenAI bits here and replace them with open source equivalents? And yeah, the answer is totally. We just need something that converts speech to text, something that goes from text to speech, and then something that does chat completion. So that's something we'll revisit a little bit later on today.
Walking through the Jupyter notebook
So that diagram we took and turned it into a Jupyter notebook that let us kind of give a proof of concept like, is this going to work? Can the model make sense of this kind of input? So I'm not going to go line by line through all of this. But I do want to show you kind of what we encountered when we saw this working. So first we load up some OpenAI credentials. We're going to make three OpenAI calls. So we need an API key. Create the client. And then the input is in the form of a video file, input.mod in this case.
So if I hit play. Can you describe what I'm wearing right now? Were you guys able to hear that?
So the first step is to take that video file and split it into audio and images. And then we're going to take that audio and we send it to the OpenAI Whisper API that does speech to text. And very quickly it comes back with, can you describe what I'm wearing right now? So perfect. That's correct.
And now we have text and we have images. So we're ready to talk to GPT 4.0. So we're going to pass in the text. We're going to pass in the images. And a system prompt. And if you're not familiar, a system prompt is like a set of instructions for the GPT to tell it what kinds of responses you're looking for.
And in this case, I want it to speak the answer. So I want it to not use sort of formal written tone, but kind of more casual tone. I want it to know that if I talk about I or me in the input, that it's talking about the person in the video. And I want to make it aware. I'm giving you images, but they're actually part of a video. So you should think of it and refer to it as such.
Question here. Where's the notebook? The notebook is in the GitHub link that is pinned above. And then Paul says, I miss what's doing the text back to speech at the end? Yeah. Well, I'll show you that next.
So let's do this completion. Takes about five seconds or so. And hey, so in the video you're wearing a gray hoodie. Great. So it worked.
And then going back from text to speech, we are using this one more open AI API. This model is called TTS1. And you can choose different voices. You do not have the option of the Scarlett Johansson voice anymore, unfortunately. But they have a number of different voices that are pretty good.
So that completed. And now in the notebook I can preview this audio. So I'm going to hit play. Hey, so in the video you're wearing a gray hoodie. The hood is down and it looks comfy and casual. Hope that helps.
One other question. So a graphics card will definitely help. For example, Lama will use an available graphics. But I believe if you're doing this, you don't need to have some sort of really fancy graphics card, just to confirm that with you guys as well. If you're going to go do something locally, then yes, you would have to have a decent one. But in this case, you do not. That's exactly right.
For this, I mean, you could probably do this on a Raspberry Pi. It's mostly making network requests. The only thing that is being computed locally is the splitting of this file. And that's pretty quick on any reasonable hardware these days.
Any other questions before we move on? Was the question about doing it with running Lama? Oh, no. We're asking, can I run this code which I don't have a good graphics card? And the answer is yes, you can do that.
You will need to install FFmpeg and you will need an OpenAI API key. And those are in the readme. Instructions for those are in the readme of the GitHub repo.
So that works. I mean, we can get a sense of how, what kind of responses we can expect. But it's a little bit tedious to do it this way.
Introducing Shiny for Python
So we can do that, but if we want to ask it question after question, that's going to be really annoying. We have to open up some kind of video recording app, save it to a file, change the file name in the Jupyter notebook, run all the cells, scroll down to the bottom, hit play. It's just going to be a really annoying workflow.
And when we're coding, it's nice to have all of that code in front of us. But when we're just trying to iterate, now the notebook is getting in our way. And, of course, if I want to share this code with other people, if I want to let non-technical people try their own queries and prompts, this is going to be extremely difficult for them as well if they don't know how to run Python.
So why don't we instead wrap all this up in a web app, and we'll use Shiny to make a much friendlier experience on top of this.
So what is Shiny? Shiny is a framework for making interactive web apps in pure Python. You do not need to know HTML, CSS, or JavaScript. If you don't want to, just pure Python, you're already halfway there if you know how to code in Python. But if you do know HTML, CSS, and JavaScript, you can absolutely use them to enhance Shiny by customizing it and extending it, which is something that I spent a lot of time doing this week, and I can't wait to show you the results in a few moments here.
Shiny is specifically designed for data-related applications. There's a lot of visualization built in that it works with Matplotlib and Plotly and is designed to work with data frames. So you can use it for a lot of different things, but it is especially well-suited for data-related applications.
The heart of Shiny is a reactive programming model. If you have not heard that term before, you don't have to worry about it right now except to know that this approach is really good at making your apps efficient because only a minimum amount of code needs to run every time you interact with your app in order to get your outputs to be correct. And it's also a very good programming paradigm for avoiding the kinds of bugs that are usually pretty common in UI programming.
The heart of Shiny is a reactive programming model. If you have not heard that term before, you don't have to worry about it right now except to know that this approach is really good at making your apps efficient because only a minimum amount of code needs to run every time you interact with your app in order to get your outputs to be correct.
It is an open-sourced framework, MIT-licensed, and after you've written your Shiny app, you can either run them locally or you can deploy them in a variety of ways in the cloud or on-prem. We have a number of free and paid cloud options as well as open-source and enterprise on-prem options, and all that is described in the Shiny for Python documentation I'll link to at the end.
Do you want to decode on-prem for those people? Oh, I'm sorry. On-prem means on-premises, meaning you're running it on your own server that you manage, whether that is in your bedroom or in your data center or in a cloud of your choice. But if you're managing the server, then that's called on-prem, whereas the cloud offerings, you don't even know that a server exists.
Jeff says, I built this Shiny app last year and made it with image capabilities too. That's a sweet feature. Nice. Thanks, Jeff.
Shiny website tour and Q&A
So this is the Shiny for Python website. We have a tutorial here, as you would expect, that leads you through many, many, many different features of Shiny and starts you all the way from zero to building very sophisticated web apps.
We've got a gallery here of the different kinds of input and output components you can put into your Shiny apps, as well as different types of layout components you can have, like navbars and sidebars and tabs. And if you click on any of these, you'll get documentation and a live example running in the browser.
And we also have templates that put together these components and layouts into sort of ready-made apps that are designed for you to kind of fork and make your own. So you can copy these shell commands to bring these projects local, and then you plug in your own data or whatever.
Question from someone. What are the free cloud options? Yeah. So we talk about that here under the Deploy tab. We have our own server called Shiny Apps IO. And I think if you use the free tier, then I believe you're limited to five applications. And there's some limits to how much scalability you get. And I think they can't be password protected or something like that. So there's some features that are held back. And then there are various paid tiers that unlock those different features.
There's also HuggingFace. HuggingFace Spaces is their sort of framework agnostic hosting option. And if you go to HuggingFace Spaces and create a new space, one of the options under Docker is Shiny for Python. So that's another option if you're familiar with HuggingFace.
Shiny versus Streamlit
A couple more questions about Shiny. What do you guys think between Shiny versus Streamlet app? Which one is more popular when it comes to build your own data app? Thank you. How is it different from Streamlet? Yeah. In terms of popularity, Streamlet is a monster.
Streamlet is by far the best known Python web application framework for data scientists. That being said, I think the reason for Streamlet's popularity is extreme ease of getting started. And I think one of the things that differentiates Shiny from Streamlet, probably the main thing, is that it is almost as easy to get started with, but it is much more powerful and flexible.
So if you know anyone who's built a lot of Streamlet apps, you can ask them. They'll probably tell you that at a certain point, you kind of need to do something else. The main reason is because every time you do anything in a Streamlet app, it re-executes your entire Streamlet app from top to bottom. And for simple apps, that's fine. And for some of the kinds of apps I'll show you today, that is very not fine.
And Shiny is designed to work from very, very simple, you know, just one or two inputs and outputs to very, very complex, dozens of pages with hundreds of outputs. And Shiny still works extremely well and efficiently.
And Shiny is designed to work from very, very simple, you know, just one or two inputs and outputs to very, very complex, dozens of pages with hundreds of outputs. And Shiny still works extremely well and efficiently.
Okay. So we do need to move on. Really, there's a lot of Shiny questions, which is really great. So Winston's going to jump in and actually answer them. And also, there will be a Q&A afterwards. So for anybody who actually wants to build this application and other apps, Joe has already agreed to kindly stay a little bit after, and he can actually show you how to build it.
And also, that question about Streamlet, I think you will realize that as we go through the rest of this project, because it showcases things. Like when we're switching models and stuff like that, there's, like, certain things that cannot be done using just Streamlet. So you can see that.
And then just one last question from Sylvia Seto about Shiny. I'm just a fool trying to run Shiny. Where do I start? Yeah. I would go to the shiny.posit.co. And just start at Learn Shiny. The first page is designed for you to just start from nothing. And, in fact, you don't even need to install Shiny right away. If you don't want to, you can pip install Shiny. But if you just want to start kicking the tires right away, right from within the web page, you can run Shiny apps, like right here.
Another option for free deployment is using Docker with GitHub Action. Sure. Yeah. If you're familiar with those things.
Shiny demo: dashboards and hand tracking
So let me just show you really quick. This is a pretty standard type of Shiny app. A lot of people use Shiny for various kinds of dashboards, BI dashboards or experiment results, things like that. And, you know, you can change parameters on the page, and everything updates, you know, right away.
So the nice thing about Shiny is that it only executes the minimum amount of code. So let's say if I uncheck this and only this table was affected by that particular control, then that's the only thing that would execute. You don't have to re-execute everything.
So let me show one more demo, and then we'll get back to GPT-4. So this is a fun demo that shows a 3D visualization of Brownian motion, which is the kind of motion that a molecule undergoes while it's sitting on a surface. So I can hit new data and randomly generate data. And some of these are pretty interesting, but then like other ones of them, like it's a little hard to see what's going on here just because of the camera angle in this 3D perspective. This is using a Plotly output, and one of the features of Plotly is that you can click and drag to move it around. So that's nice.
But to make this demo extra fun, I have added a hand tracking component. So I can just wave my hand and the plot moves. So what's happening here is that in the browser, it's using a Google library called MediaPipe to recognize the features of my hand and sending those to Shiny. And Shiny is calculating a normal vector shooting out of my palm, and then I'm using that to orient the Plotly camera. So I don't know how often you'll need to use that for your visualizations, but it's a fun little demo.
So that's Shiny in a nutshell. Now let's get to the live coding part where we'll find out how good I actually am at writing Shiny code. The guy who made Shiny.
So just as a reminder, we started with this Jupyter notebook that works, but it's a little bit clunky. The first thing I want to do is I want to take all this logic and I want to reuse it.
I want to turn it into a Python function. And luckily, Jupyter has a handy utility built in called nbconvert. And I can say, take this notebook and turn it into a Python script. And just with that one command, it gives me a Python file.
I'm going to rename it so it's easier to import. I can't import it if it has dashes. So I'll call it query.py.
And as you can see, this is Python code. It took all the cells out. So in order to use this as a function, there's a couple of things we're going to need to do. First of all, we don't want to use this hard-coded input.mov. We want to take the video file as a parameter. So let's make a function called query. And this will be the sole parameter. And all the rest of the code becomes the body of this function.
And now the only other change I need to do is go through, and there are a bunch of things I put in just to show things in the notebook. I can take them out now because we're not in a notebook anymore. But all the logic is going to remain the same.
Okay, and at the end here, this was me showing the little audio player so we could hear the response. We don't need that. Instead, we're going to take that response audio and just return it.
Building the Shiny app
Okay, so now we have a function called query sitting in a module called query. And this is not the Shiny app yet. This is just a regular Python module. And now we're going to write a Shiny app that is going to use that.
Okay, so Shiny apps are generally called app.py. And you do not author them in Jupyter Notebooks. You do them in Python files, usually using an IDE like VS Code or PyCharm.
I'm going to go very slowly for this part. The actual app is not going to end up being that big, so we're going to sort of iterate our way into it so it doesn't feel like we just sort of jumped to the end. So first, we need to import some stuff.
Are you going to say import from? It's so much harder when you know people are watching. There's a lot of expectations as well for yourself.
Okay, so I'm going to start by just putting nothing but a string there and showing that we have a Shiny VS Code extension here. And if you have that installed, then you can just hit this play button if you're on a file called app.py and using Shiny. And when you do that, it'll automatically start Shiny, and it will put a preview up here. And if I change this and hit save, then it'll automatically reload.
So like I said, we're going to do this in sort of baby steps. The first version of this, let's still use video files. I can use Shiny to let you upload.
So I'm making this function call called UI.inputFile, and that will create a file upload control. So I can say browse and then select my movie, and it'll upload it to Shiny. So far, we're not doing anything with it. So let's show something very simple about this video clip. Let's see how many bytes it has or something like that, or maybe the name of the file.
So I can access this input by using the keyword input, and then this first parameter is clip. That's the ID. So I refer to it here, and then I call it like a function to get the value out. And in this particular case, I happen to know this type of input gives me back a number of different files, because I could upload a bunch of files if I want to. So I'm just going to take the first one, and I'm going to print the name of it.
Don't worry about that error. We'll come back to it. Okay, so input.mov. And if I were to change this, then this automatically changes.
Understanding Shiny reactivity
So what is this? What is this little function with this render express on top? This is that reactivity that I was talking about. Basically, things that appear at the top level, these are things that are rendered into our web page when it first loads. And then anything that needs to change, anything that needs to re-render or recalculate over and over, you put into little functions and then decorate them with these expressions like render.express.
In this particular case, we are reading this input clip and getting the name off of the first entry. So it automatically knows that any time input.clip changes, meaning any time I upload a new clip, it will automatically re-execute this. But if I have other controls on the page and other of these little outputs that are reading those other controls, then they wouldn't need to change when the clip changes.
Basically, things that appear at the top level, these are things that are rendered into our web page when it first loads. And then anything that needs to change, anything that needs to re-render or recalculate over and over, you put into little functions and then decorate them with these expressions like render.express.
Let's do something more interesting, though. So here is our video file. This is how we actually get the data from this uploaded file. So we have a video file here. And what are we going to do? Let's display this as a video tag in... Oh, no, sorry. I forgot what we were doing. We are querying chat GPT.
Okay. So I'm going to call query and give it the video file. And that will give me an audio file as a response, right? Or audio URI. And I can take that audio URI and insert an HTML tag called audio. And this should be all I need to do to... Yeah, I think that should be it.
Okay. So I'm taking this file. I'm using this function that we ported from the notebook. That will take a while. So actually, because it's going to take a while, let's say, like, calling GPT just so that we know that it's working.
Sorry, Joe. I'm so sorry. I just need to interject. Everybody is so impressed with your abilities right now. Like, the chat is going crazy. She's actually typing things.
Oh, and by the way, because this is... This error, this is happening because input.clip. Before I've uploaded something, it's null. So I can say if input.clip is not none... It is none. Then just don't do anything. Okay. So that error is now gone. So let's try this.
Calling GPT. So it is working. And there, it worked. And let me see. Oh, because of the screen sharing, you won't be able to hear this, but I'll tell you what it says. Hey, there. So in the video, you're wearing a gray hoodie. Looks comfy. Okay. So that worked.
Let me just prove it to you. I will reshare this. Yeah, because of this, the live streaming platform, I can only share audio with you when it's a specific browser tab and not when it's the IDE. So we'll have to switch back and forth a few times.
Okay. So in the video, you're wearing a gray hoodie. Okay. Great. So that worked. That's already better-ish. I mean, we still have to record these videos, which is pretty annoying. But before we fix that, there's something else I want to fix first.
Adding progress indication
It drives me crazy that it sits there for so long showing nothing. And as a user, we're going to sit here and wonder, is this thing even working? So what we want to do is we should add, instead of printing this to the Python console, which is not going to help anyone, we can add some progress indication. So this is a feature of Shiny where we can say pset a message processing.
So everything that is inside of this with statement is going to have a progress message showing. And it's going to show processing, and that's it. And so when this line finishes executing, then the progress will automatically go away. So let's give that a try.
Yeah. So there is our progress. And now at least we have some idea that something's happening.
Replacing file upload with live video recording
Okay. So now let's get fancy. The worst part of this experience is having to upload a video clip. I don't want to fire up QuickTime Recorder every time I want to ask a new question. So I did some work in advance after Tina posed this challenge to us last week. This whole week we've been working furiously to create a couple of components that would really help this experience.
So input video clip. So I've made a dedicated control to help make this easier. So we're going to call it clip still. And it actually, instead of uploading a file, it returns data that's immediately ready to drop into the query function. It's going to be a video URI.
So let me hit save. And unfortunately, VS Code will not let you use the camera on your phone. So instead I will. Yeah. Okay. So now we have this experience. I have a webcam that's live here. I can switch to different devices. And I can hit record to start recording and then stop to stop recording.
Before I try this, let me make one more change before I switch to screen sharing so you can hear my audio. This is a little wide. So let's, because I know a lot about web development, I can say max width is 600 pixels. And it looks like this. Oh, but I want it centered and maybe with a little bit of breathing room on top. So I can, again, using my extensive web development skills, I can say put some padding in my direction and center it.
Okay. Let me now change my screen sharing so you'll be able to actually hear this.
Okay. So let's try this. Where am I right now? Hey, looking at the video, it seems like you're in an office or co-working space. The setup with the desks, partitions, and lighting gives off that vibe. It looks like a pretty productive spot. I don't know about productive, but it is a co-working space. So I got that correct. And this is so much easier for us to iterate.
I can now ask, what color is my hair? Hey, from what I can see in the video, your hair looks dark. Probably black or dark brown. Okay. Great. So we're already making a lot of progress.
Replacing the audio player with a custom component
But now I was pretty dissatisfied with the way that output control looks, like that kind of stock HTML audio player. So let's do... I created another component called audio spinner. And instead of using default HTML audio tag, I'm going to use this control.
Okay. I'm sorry. I'm going to have to do this a couple more times. Okay. What am I wearing? Hey there. In the video, you're wearing a light gray jacket over what looks like a dark gray or black shirt. And you also have white earphones in. Looking good. Great.
Okay. So now I think we are almost at the point where we can ask it any scenario that we can think of. Oh, I just realized the time. Tina, do we have a hard stop on the hour? Should I move on to the playing around with this? We don't have a hard stop. But I know some people do drop at the hour mark. So, yeah, maybe we can try out the demo. And then, yeah.
Want to show your fancy thing, though? I was very impressed.
I think it was just the fancy bit left, right? The rest of it. Oh, yeah. I showed that spinner. But let me let you... This is a little bit scary. But I'm going to let you all try this at the same time.
Oh, shit, guys. Let's not betray his trust. All right? So I've deployed this. I've not load tested it. But I'm putting it in a chat right now. Why don't you all click on that and you give this a try. Ask it questions.
I've only tested it with English. I don't know if it works with other languages. But let's give it a try. If something funny happens or something interesting, then post it in the chat. Don't spam, okay? I was nervous right now, guys. We make a good impression. Be on your best behavior. Stop doing strange things.
This guy is crazy. You don't know how crazy. This is my personal open AI API key. Oh, my god. Oh, my god. We'll see. We'll see.
Wait. So what's the link supposed to be? Oh, yeah. You can't paste the link directly. Can you paste it on the reverse side? I'll paste it for you.
This must be Joe's first time on the internet. There's the link.
Whilst you're doing that. You guys do that, and then we'll test the demos. Do you want to, yeah. Am I going to do it, or do you want to do it using that link? Oh. Oh, we can do that. Actually, Winston, would you like to do it? Let me see if I can make this work.
Okay. Let me. All right. Share. Oh, man. I'm not dressed for this. I don't know what that means, but that doesn't sound good.
All right. So here I am sharing. Can you all see this? Yep. Okay. Great. All right. Somebody has to tell me something interesting to ask it. Oh, no. We have the demos. We have the demos here.
So, okay. Actually, first question. Do you speak another language? I do, but I speak a language that it probably doesn't know. So I speak some Taiwanese. It probably does not know. Okay. All right. Maybe we'll try that one in a bit. I can do that one.
Why don't we do the, so here. Like, Winston, do you mind? Put it here. I'm just sharing the thing over here. Sorry, which thing? I have to put it in Riverside. So we're going to replicate this demo. Can you just do that? Oh, okay. Okay. Got it. Sure.
What the heck is this? This is the demo from OpenAI. Okay. Here we go. Were you able to hear that? No. Oh, my gosh. Then you need to share. Here, I'll do it. Don't worry. I'll share the tab for you guys. Do not worry. I'm saying that because I'm worried. I'm like, don't worry. Don't worry.
I've seen the chat. It seems like people are trying it, and it's working, which is great to hear.
Live demo: testing the AI assistant
So this part. I want you to tell them a bedtime story about robots and love. Oh, a bedtime story about robots. Okay. So we're going to do that, and then I will go and.
Okay. I think you guys can see that, right? It worked with French. It worked with French.
Tell me a story about robots and love. I also might not be the best person to do this just because I do not have the best internet right now, but let's see if this one will work. Sure. I'd love to tell you a story about robots and love. Once upon a time, in a future not too far away, there were two robots, Luna and Sol, who worked at a bustling city's central energy plant.
Oh, it looks like you're reporting for NBC News, huh?
Okay, let me just refresh it. So, Joe, this version of it doesn't have the history? It doesn't have history, so I think I need to, like. No, I think I meant to deploy the one with history. Yeah. Does it have history? It's supposed to. Okay. It might be my. I'm not juggling any versions of this app today, so let me see. Yeah, this is the one with history. It does have history? Okay, it might be my internet as well. It's just because it's shitty Hong Kong hotel internet right now. But I'm going to try.
Can you tell me a story about robots and love, but make it really, really, really dramatic?
The longer the answers are, the longer it takes both to generate and synthesize. I'm going to pass it on to you after this. Winston, if you don't mind, you can do the one. Actually, this one. Yeah, you can. I just think my thing is so slow at this point. I'll send you the links. Oh, sure.
Here it goes. In a future where robots and humans coexisted, there lived a robot named Aria, designed with the ability to feel emotions. Aria served a wealthy family, but deep down, she yearned to experience love. One day, she met another robot named Pax while recharging at a robot station.
Okay, that's great, but more emotions. More emotions in your voice. That's not going to work, right? No. It won't work. Because you need that integration. The part that understands what you mean by more emotion is not the same that's transcribing the text into speech. So, yeah, there's definitely some of that.
Okay, okay. Yeah. So, do you think there's a way for us to actually vary the emotional level of the voice? Because I think they were trying to show the fact.
The longer the answers are, the longer it takes both to generate and synthesize.
Oh. Lost your screen. Blinked out for a second. Lost you for a second. Oh, no, I can't hear you. Oh, got it. All right, here goes a more emotional take. In a far-off future where robots and humans coexisted, there was a robot named Aria.
I have not had that happen before, actually. Let me figure it out. Wow, the robot overlords. Oh, no. Okay, let me see. Yeah, it says cannot connect streaming software. Viewers will be able to. What happened? Oh, wait. Are we back? No, it's back. It's back. Okay, okay, great. Okay, let's just pretend that. Okay, cool. Hi, everybody. We're just trying to figure out what happened. My camera has decided to lose focus.
All right. It's a live demo, right? And we also didn't rehearse this. No, I had no idea, actually, that we were going to do this. Yeah, exactly.
I've got a paper here. I've got a paper here. I'm going to write down an equation and ask it. What was I supposed to ask it? I'll play the demo. So you're asking, can you help me solve this? Can you help me solve this math equation? Just write 3X. Yeah, yeah.
Okay, so here we go. Where's my mouse? Here we go. Can you help me solve this math equation that you see here? It is reversed, so I'm curious. Sure thing. So in the video, I see the equation is 3X plus 1 equals 4. Let's solve it step by step. First, let's subtract 1 from both sides of the equation. 3X1 minus 1 equals 4 minus 1. This simplifies to 3X equals 3. Now we divide both sides by 3. 3X squared 3 equals 3, 3. So X equals 1. There you go. The value of X is 1.
Wow. That's impressive. I'm surprised. I'm surprised. You know, I've tried to use ChatGPT to do other more sophisticated math, and I'm a little afraid to try that because it has failed miserably. But that was pretty cool. Do it. Do it. Do it. Write some fancy math. No. You guys got to help me out here because coming up with that off the top of my head is a little bit.
All right. Everybody also write in the chat, and then we can test things out as well.
Exploring multimodal features
In the meantime, for the next one is the code. So what they did have, what they demoed there was that they were able to take the code, and there's like a function on the app itself where you can just copy-paste it into the UI itself. Oh, you can do that? No, I didn't think of that. Okay. But you could have done it. I could have done it. That part of the demo, yeah. I could have done it. I just didn't. There you go.
Everybody's going like, Joe is insane. Yeah. So that part, but what we can do is we can try showing it the code that you wrote and then ask you to interpret it and then show it a graph as well. Let me just pull up some of this code here. And I do have that other open source comparison that we should hit as well.
Okay. All right. Let's move. Oh, sorry. What do you guys want to do here? Do you want to just do that one real quick? And then that's why don't we just, I'll do the language. Wait, Joe, do you speak another language? No. Well, not well enough that I'd want to do it in front of you. Okay. Because my Chinese is kind of shit right now. But okay, I'll do that one.
All right. All right. Your Chinese is fine. I think you should do that. Okay. So that one. And then I think let's also do the one with your facial expression. Okay. I'll do the Chinese one real quick. And then maybe Joe or Winston, one of you guys do the other one. Okay. Internet, you can do this. You can do this.
All right. Share my screen. I am here.
Wait, what did, what do I say?
Maybe that was really long. I don't think it's. You're like, is it working? Yeah.
Somebody said they tried it with Arabic. It's working. Wow. It's working. I think you should go for something light and delicious. Maybe some dumplings or a bowl of noodle soup. Pretty good. If you're in the mood for something sweet, how about some fruit or a pastry? Just pick something that makes you happy and enjoy it. Oh, thank you. Nice. That was pretty good. That was really good.
I was going to ask it to respond in English, but I asked it what should I do after this live stream? What should I go eat? So that was really good. All right. You want to do the facial expression one, Joe? Yeah. Am I just asking it what kind of facial expression am I making? Yeah, but do like an expression.
Yeah. Okay, I'm going to do it with this special version that you have not seen yet. Oh, wow.
So Tina raised the question of how does this model do versus open source multimodal models. So this version of the app is going to run it against three models simultaneously. So the new GPT 4.0, the existing GPT 4 Vision Preview, and this open source one called Lava.
So what kind of facial expression am I making here?
Okay, so it's asking all three.
Okay. So we can try these as they come in. All right. Let's check out your expressions in the video. It looks like you're making a few different faces here. First, it seems like you're concentrating on something or listening intently. Then you shift to a kind of playful or inquisitive expression. Towards the end, it looks like you're squinting or maybe trying to figure something out. Super expressive.
So that's one thing about this, too, is that the speech and the video is not lined up. So it doesn't know when I finish asking the question, then I made a face. It doesn't know that. It just has a series of images. So I wonder if it will do better. So I wonder if it will do better when the real API is available.
Let's see how GPT 4 did. It looks like you're cycling through a few different expressions here. At the start, you seem to have a neutral or slightly serious look. Then your expression shifts to a small smile, and then it transitions into a more pronounced smile. Almost as if you're about to laugh. Lastly, your expression changes to a slight frown with raised eyebrows, giving off a look of confusion or concern. It's like you're reacting to different things happening, each requiring a different facial response.
I think this one's better. I'm going to vote for that. It looks like the person in the video is making a surprised or curious facial expression. They seem to be looking directly at something, possibly reacting to someone or something off camera.
Okay, well that was... I think GPT 4.0 is the best. So I voted for it, and then it goes on to this leaderboard visualization here. So I can keep the conversation going. This time I'll make a consistent face the whole time. So I'll go like... How about now?
Alright, let's see what's going on in these frames. It looks like in the video you're making an expression that seems a bit concerned or worried. Your eyebrows are raised, and your mouth is kind of downturned, which usually indicates concern or maybe a bit of confusion. It's like you're reacting to something unexpected or trying to figure something out. Super expressive.
In this video, you're showing an expression that suggests concern or sadness. Your eyebrows are raised and pulled together, and your mouth is slightly turned down at the corners. This type of expression might indicate that you're feeling a bit worried, unsure, or possibly empathetic to something that you've heard or seen.
Hey there, how can I assist you today?
Okay, so not the open source one. I think I have to give to this one again. So you get the idea. You can keep doing this over and over, and you can keep the score of who's doing the best here. And if I had more time, we could do this in a way that everybody could participate, and this would be the global leaderboard. But that would require a database, which I didn't feel like setting up.
Running Lava locally with Ollama
So someone asked, where are you running Lava? And because I'm running it locally, I can't deploy it. I never used Lava until yesterday, basically. And I did it with Ollama, which I only know how to run locally right now. So I'll deploy it if I can figure out a way to do it in a sensible way. But that's the reason this is just local.
If you want, this version is in the repo. It's called app-bakeoff. So if you load that up... There's actually instructions in the readme of how to run this app. It's the very last thing in the readme is how to run this app locally. So you can try it if you want. But you need to have Ollama installed as well. Installing Ollama is super, super easy. When I did it, I was shocked at how easy it was. So you guys can try it out.
Trying suggestions from comments
Do you guys want to do maybe, if you don't mind, a couple of the suggestions from comments? And then we can go into Q&A. And then after that, maybe, Joe, you can show some of the other shiny apps. I think people had a bunch of questions earlier about how to build things.
I can do the math equation. Let me just write that down here. The more complicated one. The more complicated one? x plus 2y equals 4. And x plus y equals 5. My experience is that ChatGPT is not very good at math.
I'm also curious. Can we get it to actually speak in another language? I want to see if it's good at actually speaking the language. Can you please solve the math equation that's written here? Sorry, the pair of math equations that are written here.
Sure thing. Let's solve the pair of equations written in the video. Here are the equations. 1, 2x plus 2y equals 4. 2, x plus y equals 5. First, let's solve the second equation for y. x plus y equals 5 implies y equal 5 minus x. Now, let's substitute y in the first equation. 2x plus 2, 5 minus x. This simplifies to 2x plus 10 minus 2x equals 4 in place 10 equals 4. Oops, that means there's no solution to this system because it leads to a contradiction. Equations represent parallel lines that never intersect. So, no values of x and y will satisfy both equations at the same time.
I'm having trouble thinking in my feet. Is that right? You do not know the answer. I'm sorry, my brain is like... I'm going to stop struggling before I embarrass myself some more here. No, please don't.
I am actually pretty good at math.
Does anybody know the answer? In the meantime, I do want to answer this one question, which I think is really important. It's basically saying this can be really scary because it can be used in problematic ways when it comes to programs dealing with humans and especially without proper supervision. Honestly, I absolutely agree. That's something that I brought up a lot in videos as well.
Responsible AI and Posit's open source values
One thing I want to note is that technology is not inherently good or bad. This technology, I think the one thing we can agree on is that it is going to rapidly develop whether we like it or not. But it is up to the responsibility of all of us in order to actually make sure it's in the right hands doing the right things.
I actually think, if you don't mind, Joe or Winston, I think this is actually a great place to talk about what Posit believes in. Because Posit has a very special strong stance surrounding open source in these topics. I'm actually going to let you guys talk about that.
I think we do have a very strong stance around open source and less directly relevant to LLMs. But our general philosophy as a company is we believe in developing tools that help people understand the world using data and making that available to everybody regardless of economic means, which is why all of our core technology is open source.
That being said, when it comes to these sort of large AI models that we don't really fully understand and don't have an accurate mental model of how they work and how we can sort of keep them safe, I think within the company there's a lot of different opinions. And I think some people are very, very optimistic and some people are very, very pessimistic.
And I think there are people in the company that are working on taking this technology and making it really good for our users by enhancing the experience in our studio and letting people build tools in Shiny that help enhance their experience of working with these AIs. But we also have people at and affiliated with the company that are working on AI safety and are extremely, extremely concerned about what this means for humanity. And we have a lot of deep involvement in different efforts to make that safe. I think we'll be talking more about that in the future. It's nothing to share publicly right now.
Our general philosophy as a company is we believe in developing tools that help people understand the world using data and making that available to everybody regardless of economic means, which is why all of our core technology is open source.
Thank you so much. Yeah, Posit really is a very special company. There were many times in which you guys could have chosen to not be open source. There's a lot of times in which you could have not prioritized that. But every step of the way, that's what you prioritize. And I think, yeah, in my opinion, at least, it was what really stood out when I first interacted with you guys, like how strong of a stance.
And behaviorally speaking, some companies, they go like, oh, safety is important. They just leave it as that, where they're like, oh, we should think about this. But really, Posit has done that every step of the way. And yeah, I just love that about them.
Language demo and food challenge
I do. So for the last quick little demo before we go into some Q&A, I'm so sorry, guys. We are definitely over. But I want to try the language thing. And also somebody was, I really liked this one as well. Can somebody show a picture of your fridge and ask it what it can make? Does someone have a fridge?
Yeah, do either of you have a fridge? I don't have one. Bring your fridge. How am I going to do that? If I use my phone, then it won't be on the stream. Can you bring like food items, like three items? All right, I'll be back. Can you bring a laptop down to the kitchen? No, the whole setup is a little bit too complicated. All right, I'll grab something. I'll be back.
In the meantime, Joe, do you mind asking it just to say something in a different language? I'm really curious about that part.
Sorry, let me share once again. I'll go to the deployed version.
Can you say good morning to me in French? That's probably a little too simple.
Sure thing. Bonjour. That's how you say good morning in French. Have a great day. It's only a little longer.
I'm sure this is going to take a while.
I just realized I don't speak Japanese.
That sure sounds like Japanese to me. Can you choose Chinese so I can understand it? Can you tell that same story, but now in Mandarin?
Long, long time ago, there was a little village. There lived a little boy named Tai Lang. Tai Lang loved to take adventures and would go into the woods every day. One day, Tai Lang found a shining stone in the deep woods. This stone has magical power. It's said that anyone who touches it can make his wish come true. That's really good. Yeah, that's pretty cool. That's really good.
Okay, sorry. Oh wait, Winston's back. Can you ask it to say something really, really obscure, like a very obscure language? Does anybody know a language that you think that it would not know?
Well, I said I don't think it knows Taiwanese, right? Oh, okay.
Can you tell that story again, but this time in Taiwanese?
Hey, no problem. This story goes like this. Long, long time ago, there was a little village. There lived a little boy named Tai Lang.
Oh, wait a minute, wait a minute. Well, you didn't specify the dialect, right? Did you? I said Taiwanese. Is that... Yeah, I don't know. Why didn't that... Do you want to do Moroccan Arabic? Oh, I was trying. All right. Ask for Klingon. Klingon.
Hindi. That's Pronto. That's Pronto. That's a good one.
All right. Here's a short story in Klingon for you.
Sounds like it's like reading English, like Klingon written out in English, as opposed to American English or something. Yeah. Maybe the text-to-speech is not trained on that, but like the model is. Yeah, the accent, I feel like the Taiwanese, I can tell. It sounded like pretty mainland Chinese to me. Yeah.
Sorry. I need to choose another. Can you try Esperanto? No one knows that, not even me. Okay, something that people actually know.
Hungarian. Do you want to do Hungarian? Now translate it to Hungarian.
Sure. Here's the story translated into Hungarian. Köszönöm, hogy megvilágosítottál, vezércsillagom. A hajó úgy haladt át a végtelenségen, mint egy hulló csillag, és amikor elérte a ködöt, a hajó fegyverei hírtottak jelenlétéről. A siker a te szívedben is ott volt, ezért hétméltatlan ellenség egyesült, hogy tanulja legyen erejének. Így megtanított nekem a kardútját is. Remélem tetszett.
Köszönöm, hogy megvilágosítottál, vezércsillagom. A hajó úgy haladt át a végtelenségen, mint egy hulló csillag, és amikor elérte a ködöt, a hajó fegyverei hírtottak jelenlétéről.
Oké, itt van egy kicsit a főterem, mit kell tennem nekik? Nem fognak jók lenni. Bármi, amit mondhatom, hogy nekem tennem, egy kis frutokat és egy kis mañával, nem fognak jók lenni. Köszönöm.
Ez nagyon jó. Kérdezd meg a hindihoz. Remélem, hogy hindi is ismer. Igen. Nagyon sok ember ezt beszél.
A hangulás, a hangulás hangulás, nagyon különleges volt. A hangulás, a hangulás, nagyon különleges volt.
Jó. Jó. Nem tudom hindi, de... Vagyis valaki a csatornának lehet elmondani, hogy ez valószínű. Oké, úgyhogy, hogy mindent befejezzünk. Először, minden linkset, és ha már nem voltál, kérlek csináld ebben a híváson. Csak hétfőn. Visszajelződök neked az állatokat, minden hétfőn, és aztán mondom neked, hogy mi a következő hangulás. Ha ezt csinálsz, meg tudom visszajelzni neked.
Majó és szaladás. Ez valószínűleg nem túl rossz volt. Amit gondoltam, hogy meg kell csinálnod egy szaladást. Majó és szaladás szaladást.
De a szaladás... Szerintem sokáig érdekel, hogy milyen jól tűnik. Nem gondolod, hogy jól tűnik, vagy nem jól tűnik? Nem tudom ezt. Nem érzem meg, hogy jól tűnik. Csak maradsz ott, megkóstolásokat és majonásokat, és úgy gondolod, igen, jól tűnik.
Shiny app categories and templates
Jó. Meg tudom kinyitni ezeket. Ezek egyszerűen egyszerű... Egy egyszerűen egyszerű kategória az appok. Ez azon volt, amikor Chinese 2012-ben kezdett, hogy egy professzor volt, aki akarta, hogy legyen egy ilyen interaktív kérdés a különböző statisztikai konceptekre. Ez nem egy nagyon jó példa, de csak ezt a formát, ahol egy visualizáció van, és kétféle másik módon tudod kinyitni.
Ha ez a visuális forma az, amit akarod, akkor ez a lisztet, vagy a fóluskat. Viszont nagyon közel vagy sokan uramodó kis algoritmáványal, hogy legsikben íni egy olyan dasezt egy legutóbbiokra. Tudod, az a rendek, hogy kéne reményítni ezt a kurvázást, hogy ettől nem fogsz mennyit kicsivázni. Így ez a különböző, vagy kien bármilyen átlátó bentolyat darab. Különböző alakokat tudod elválasztani. Vagy különböző módokat elválasztani a lakókat.
A második kategóriája azok a dashboardok. Úgy gondolkodik, hogy a logika ezekben rendben állító. Csak dataframe-menipulációt csinálod PANVAS-on, és visualizálod valamit Plotly-től vagy MapPlotLib-től.
Ha nem ismered Shiny-től, az állapot a különböző része. Ahol tudod, hogy ezt akarod látni, de nem tudod, hogy hogy. Ezért készültünk ezeket a templeteket, hogy megjelenítsd. Ha nézel a kódjára, láthatod, hogy az állapot itt van. Ez az, hogy különböző alakokat ad, és minden változatokban egy funkciójában van. Meg tudod változtatni ezeket a saját változatokban, a saját datacerteket, a saját különböző alakokat, és olyan dolgokat.
Shiny Core vs Shiny Express
Egy dolog, amit nem számítottam, de nagyon fontos, és a tutóriálon megjelentünk, az, hogy van két verszió a Shiny-től, vagy két mód, hogy készítsd a Shiny-től. A Shiny Core és a Shiny Express. Minden, amit most csináltam, a Shiny Express volt, amikor jelentettünk, hogy készítsd.
Tehát, amikor a templet különböző részekben van, ha csak elkezdted, szeretném, hogy állítsd ki az App-Xpress versziót, és aztán az egész azzal a saját különböző részekben jelentős részekben is, és ez csak egy másik mód, hogy készítsd a Shiny-től, amikor minden a UI-t egy nagy kis része alatt jelenti, és aztán minden a dinamikus különböző változatok egy nagy kis része alatt jelenti.
Tehát, ha tetszett, hogy a dolgok egyébként különböződnek, akkor ez az, hogy a Shiny-nél a különböző része, és aztán ez a Express-ság minden egyáltalán egymással megjelenti, mint egy normális Python-társaság.
Igen, és nem is elkezdtünk ezt elkezdeni, de sokan, amit is a saját különböző részekben készítek, azokat a data-teljesítményekre. Például, mi vannak olyan appok, mint ez, hogy minden évben egy konferencia van, ahol 1-2 ezer ember jöjjön, és 100-100 különböző különböző részek jelentkeznek, és a családban az emberek különböző különböző részeket készítenek, és megjelenik a különböző részek különböző részek különböző részeket a data-teljesítményekre, és mi a Shiny-különböző appokat készítünk a data-teljesítményekre.
Shiny Live and running Python in the browser
és van két másik hely, ahol van egy kicsit egyszerű események, van a gáború, ezek egy kicsit felváltóak, de van egy kicsit jók, majd van egy játékpont, amit nem értem ebben, de a Shiny játszik Python-t mindig, amikor változtatod a különböző részeket, mindig, amikor játszol a különböző appon, úgyhogy mindig Python-t kell játszani a technológiával Shiny Live-on, ahol gyakorlatilag Python-t játszunk a különböző részekben, úgyhogy minden ezek az applikációk nincs különböző része, minden a különböző részekben játszik, például Python a különböző részekben játszik. Úgyhogy lehet nézni ezeket az applikációkat, és lehet változtatni őket a különböző részekben, és játszani, és újra, ez teljesen
jó a különböző részekben, de ez nem teljesen erősebb, tehát ez egy különböző részekben játszani, és ez egy erősebb részekben játszani, és ez egy erősebb részekben játszani, és amit a különböző részekben játszani azokon
nem ma, és ez kicsit kicsit nehéz, mert ez Python, szóval azzal a lokalitással a Android-on, nem tudom milyen működik, de azt mondom, hogy a web működik, hogy a web működik, de a web működik.
de a web működik, de a web működik a különböző szent nem az egy egy egy egy egy egy egy egy egy egy egy egy egy egy egy egy egy egy egy egy
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
d d d d d d d d d d d d d d d d d d d d d d d d d d d d d d d d d d d d d d d d d d d d d d d d d d d d d d d d d d d d d d d d d d