Let's "Imagine with Claude" live

**[00:00:00]** [Music] All right. Hello everyone. My name is Arando Prescareno and today we are going to be talking about Claude Sonet 4.5 and a few other things as well. I'm just checking the screen here to see that we are streaming to the different platforms. So you can catch us on YouTube x LinkedIn and Instagram apparently is streaming as well. So, if you didn't know, if you use Riverside, we use Riverside specifically for our podcast recordings, but you can also use it for live streaming like I'm doing right now, and it has a lot of destinations like YouTube X, LinkedIn, Facebook, etc. I've never gotten Facebook to work. There's some issue because I have too many pages that I manage, and it just the list is not long enough. Um, but Instagram is not on the list. However, they do have a custom RTMP option. So, if you go to Instagram and start a live stream, you can just grab that URL and key. And looks like it's coming through. So, awesome. Let me know if you hear me fine. And if you have any questions, please drop them in the chat as well. So, what we're going to be talking about today is a lot of fun because Cloud Code, if you don't know what it is, is an amazing tool for either software developers or even marketers or just even just a regular business owner or a person who just wants to build things. So, if you have any questions, drop them in the chat. There's there's so much to talk about, but I'm just going to talk about specifically the new update first. So Claude is like a chat GPT. For those of you who have no idea what I'm talking about, Claude is **[00:02:00]** Enthropic, which is another company, Anthropics offering for an LLM, a chatbased thing that you can talk to and it'll give you responses. It'll it'll help you out, right? So Anthropics Claude is very good. You can use it in the browser. You can use it on your phone, but you can also use it in your console in your terminal if you are developing something. So there's a cloud code which is like their CLI versions, command line interface version that developers really love because it's very powerful and allows you to do a lot of things. Yesterday I was talking about how I told the client, hey yeah, if you want to rebuild your app, he just consider this approach just as you're prototyping and just thinking about it and I showed him what we could do with cloud code just actually just prompting it and saying I want this interface and I showed it some screenshots and it would go along and actually create things and it got very far. So it's very useful. We use it a lot in development. And obviously you want to know what you're doing. If you're doing anything in production or anything that's mission critical. But if let's say you are a software engineer and you spend a lot of time on trivial coding tasks, these tools help a lot because they know the languages very well. So I'm talking about programming languages here. And what this means is that if you know what you're going to code and you know the important part of it. So if you know the logic that's supposed to happen, then you don't have to spend the time figuring out the syntax for that particular language because these LMS are very good at doing that **[00:04:00]** particular thing. So anyway, that aside, let's talk about the updates. So, Cloud Sonnet 4.5 just released yesterday. I was using it in the middle of the night, just the regular Cloud Sonnet. I typically move between using cloud uh within cloud code using Opus and using Sonnet. Opus was sort of their master biggest one for complex challenges. And I would find myself [Music] sometimes having I don't know if any of you have had this issue but sometimes I would be working on Opus and be thinking wow this model is amazing. It just knows what I'm asking it to do. It's very good at examining big code bases etc and would be working fantastic but then all of a sudden usage runs out. So if if you don't know you there are different tiers. There's a basic tier, right? There's the max plan and there's like a bayer max plan, but just OPUS just consumes a ton of tokens and so you would in the middle of work be sort of downgraded to Sonnet, which is actually a fantastic model anyway, but if you're used to the other one doing things right, then it felt weird. I just I just find it weird sometimes when it changes model. And so recently, I've started just I'm just going to use Sonnet and just get used to its limitations or or or how it works and and use that specifically going forward. and just for something very very complex then I would I would use Opus. And so I' I've been doing that. But then yesterday I'm always updating it just in case to see if anything new comes up. And I'm working um yesterday night, well Sunday night, right? I I was working on some things. Um just **[00:06:00]** just having some fun and on the in the morning I just felt that set just a regular set 4.1 was working pretty well. I mean I I felt it it had gotten over some recent hiccups that we have seen and I was impressed. I was like oh great but then I hear that someone mentions 4.5. What what what is 4.5? What's what's all this about? What did this come from? And so I I go and see and sure enough I do an update and I get 4.5 and it's this massive massive update that they did. It's a it's a thinking model that's very impressive and we'll look at some of the benchmarks in a second. But just out of the blue, it comes out. It is amazing. I'm going to show you what it can do. But they I think did something very well, which is they created a little playground for people to see the potential of what all this means. So that's what I'm going to talk about right now. Um, for those of you who use cloud code or or have used it before or maybe you're curious, you've heard about it, but you you haven't used it because you needed a max plan. I think now it's available in the pro plan as well. Um, I'm I'm going to show you what changed really quickly and then I'm going to go and and show some of the benchmarks and some other things as well. So, let's just make sure that the stream is working over here Perfect. All right. So, if I share my try to share cloud code. Here it is. Okay, there we go. Okay, so you should you should see cloud code here. And before when you did **[00:08:00]** slashmodel, you would get four different options. Believe it was four options, but basically it would allow you to do the default which was us which is the highest tier model. it would do ous. And sorry for those who are on Instagram because it's a vertical uh view and I don't know it's it's not showing it nicely in the vertical view. Maybe I can make it smaller like this. It will make it better. I don't know. We'll see. Anyway, it's not as important. But what I want to show is that the the default used to be opus up to like 20% of the usage and then it would drop down to sonnet. And another option was just opus and then just stop. And the other was just set. And then the last one was like Opus plan mode, which is what a lot of us did. You just used it for planning. And then uh for everything else, we just went into Sonnet for the actual building. And and some people swear by Sonnet. They just really liked it. And and as of late, I I've been using Sonnet specifically. But with this 4.5, as you saw, and of course the terminal is looking a little funny, but as you saw, now there's just two option. either default, which is 4.5, which they say is the smartest model for daily use, or Opus. There's no more back and forth, just either one or the other. So, it just simplifies it for you as well, uh, not having to think about that uh, specific thing. So, that's an an interesting thing that I saw there. But, uh, let's look at some of the benchmarks here. I'm gonna share my screen and we're going to do window and **[00:10:00]** we are going to go here and I Perfect. And I think I'm going to make myself the view toggle disappeared from. Here it is. Okay. I like this view better. Yeah, there we go. Okay. So, what we're seeing here are some of the benchmarks. So, as you can see, Sonnet 4.5 is is being shown as better than obviously Sonnet, but also Opus 4.1. Right? So the software engineering benchmarks appear to be very high, very good uh for it. And as we can see here, and I don't know if this will come through on Instagram, but if not, we're on LinkedIn, YouTube X, etc. If you see these benchmarks, they're they're pretty impressive just because they are already really good. just they were really good by default but now I mean you can see these uh are just very very good results in all sorts of different things uh reasoning graduate level reasoning high school math computer use agentic terminal use etc and in all different industries finance law medicine STEM etc it's it's also very very good compared to other things and as you can see this is something I realized this morning there's two options. So there's the regular one and there's this thinking model. So when you are in the in plot code for example, you can say turn on thinking or not. And so you can use it as a thinking model or a non-thinking model. And it's very impressive what what it can do. But uh yeah, so what I really want to showcase today and and again I'm sorry for for anyone on Instagram. I don't know if this will come through because this is a different size browser, but I want to show you something that they came up **[00:12:00]** with. I haven't tested it myself. I'm going to test it right now. And it's called Imagine with Claude. And it's a demonstration that they're making for people to see what these tools are capable of. So if you are a business owner and you just want to see what's possible and you don't want to set up the whole flow of coding etc etc this and other experiences like it I think will be very helpful for people to envision this new upand cominging future that for many it's already a present that you can just build things that you need in the moment and I was talking with Mike Carlo about this uh last time we did a live stream It's I think it's the era of I need to do something right now and I need a tool or a widget to do it and I could just make this disposable thing that I'm going to use for during the period of time that I need to use it. I can keep it of course if I want to, but I could just discard it if it's no longer necessary. So for many little things it might make sense to do things like I don't know create a little dashboard to analyze specific data just for that moment in time and then just throw it away because it's not going to be relevant in a few days or in a few weeks. So that's becomes a very interesting proposition as to where we're going with all this. What what really is the future with computers and how we interact with them. So, what I'm going to do right now, I see everything is coming through here, is I'm going to do a demo. And if you have **[00:14:00]** any suggestions about what uh what we should try, please let me know. But I'm going to jump into Imagine with Claude and we're going to see what it can or can't do. And maybe it'll work, maybe it'll break. Have no idea. But, uh, let's go. So, what I'm going to do now is going to share my screen. I'm going to open this one and let's click. Let's go. Whoa. Had a nice little intro thing. Okay. So, note to self. We've got these. We've got cloud code. Oh, it's just like a little link to it. What do you want to build? Okay. So, choose your own adventure. Get me a dog pitch type blah blah blah. So, I don't know what should we try. I I have there's an idea that that I had that I think is useful. Or let's say you stream like like this, right? Or you have a podcast or something and you you sometimes are not sure what you wore last time. So, I've I've thought like it would be really nice to have a little app, web app where I could just take a picture of what I'm wearing today and uh tag it with something like podcast or something like that. So, so let's try that. So, the idea would be a little app that allows us to see what I wore last time. So, that way I could just ask it, okay, so what did I wear last time for the podcast? and it could show me just a grid of images of what I was wearing these past few times, right? I mean, something like that. I think that would be helpful because you could just tag it with things, basic things like um I don't know, **[00:16:00]** school, uh the kids school or the theater or uh the basketball game, I don't know. And so if you tag it with different categories, then it's very easy to just go in and check. Okay, so oh, we're going to the kids school again. Let's let me make sure I'm not wearing the same thing, right? It's going to show you like the past clothes that you wore uh or whatever hats, whatever you want. But that's that's the idea. So, let's try that one. app that allows me to take a picture of my of the clothes I'm wearing and tag event or yeah let's call event type event it is for for example I might take picture today because I am recording a podcast episode and next time I want to record an episode, I can just go in, click on the an at a glance view of the last, I don't know, few items I wore uh with dates I think something like that. Another example would be kids school so that I know what I wore that time etc. But I don't know. Let's see what it does with this. Okay, let's go again. I have no idea what this thing will do. So, so here it is. The access to my camera. Oh, okay. Uh, let's see. Yeah, sure. Allow visiting the site. Outfit tracker. All right, let's see. So, it's saying it's telling us what it's trying to do. Keep it clean. Minimal is fresh start. Shows empty state few tags. My outfit podcast kids school work meeting. It's building this in real time, right? There's user received the initial UI. Okay. Done. Context 100,000 tokens. Okay. Perfect. So, so here it is. So the way from what I understand the way this **[00:18:00]** works it it'll give you like the initial thing the first run and then as you interact with it it's supposed to build the next things. So so this this is a little crazy but let's let's see let's see what it does. Okay. So add outfit. Okay. So supposedly when you click on something, you see it says user clicked. So then it builds it's building the UI for the start camera view and everything. So it's building it as I click things. So then I see oh there I am. Okay. So okay let's see the camera has started. I need to remove the camera permissions buttons. This is no longer needed. The camera container has been updated with a shutter button. Now I should wait for the user to take a photo. System will notify me when the user clicks the shutter button. I'll receive the data. Okay, take a photo. Okay, perfect. And then the user took the photo. I should grab the blob URL and I to replace the camera view with the captured photo. Show it. The user took a photo. Now I've displayed it. Revealed the tag selection. Okay, perfect. So podcast, right? That's where we are now. Save outfit. So again, as I click it, it's doing the thing. My outfits track what you wear for different occasions. Podcast has one outfit. Kids school has another. Work meeting has another. Okay, perfect. So user saved the outfit to the podcast. I've updated a main view to show that has now one outfit. All right. So let's let's click on that one and see what happens. Podcast outfits. went out to record it and it's the picture of me September 30th blue shirt headphones. Oh wow, it analyzed the analyze a picture. That's **[00:20:00]** so cool. Okay, so it it's not only creating the UI, it's it's going a little further and and doing image analysis on it as well when when it's uh this is so cool. Okay, so interesting. Now these user clicked. So I I guess it's gonna go into the details. Yes. Blue plat shirt notes. Okay. Uh wore it when explaining um imagine with claude. Okay. Oh no, I Anyway, it'll build it'll build that part. Okay. No out record. Okay. So now I I think I'm going to ask it a few things. So here I'm going to add another outfit because I think like for for I think we need we might need some tags, right? So apart from the category. So let's say I'm here. I'm going to There we go. Take a okay. Okay. Does it realize that it can't open it Let's see if it works. because it did something with the initialization of the camera. So I don't know if it's supposed to camera. Okay, it seems the camera is already active from the previous time. The shutter button is showing. Okay, error taking photo. Okay, I am getting an alert saying camera not initialized. And let's see if Okay. Looks like it's using an MCP for the for the camera. That's There it is. Okay. So, camera started. And then can I do it now? Take photo. Okay. All right. So, I took the photo, then revealed the tax selection. Okay. and save it. Now, I for example, the shirt thing I think was a good idea, the the explanation. So perhaps it would be great if apart from right of adding notes about and style of the shirt. And this case showing headphones that those were actually tags that we can **[00:22:00]** then use to filter photos different from the main categories like podcast etc. Okay, so let's see what this does. Category tags, blue pl. There it is. Okay, excellent. And so I can add tags like daytime, I don't know, whatever. And so maybe we would do something like if you have a specific guest, right? So when's it? Yeah, daytime. There it is. And then guest uh just just because we're talking about this. So let's say I invited Mike. There it is. So that way I expect down here we should be able to see the tags Yeah. Filter by. Okay. So then I could just filter by a specific one. And then if I have multiple images, then it should be able to do that. This is impressive. I don't know what you think, but the fact that it's doing all this and it actually got access from the camera and everything, it's it's quite quite ridiculous. All right, so done context. Okay, perfect. So let's let's add another one here just just so we see it side by side because this is already useful. Like if this thing just was there, then I could just go in and see and I don't have to go into my photos and and do it again, right? So, let's start camera. And so, let's imagine now I'm Let's Yeah, let's put this up here like this. Let's see if it does something different. And perfect. I can't see the button because this Perfect. And so here I can then say if I if I want to filter by different things. Okay. It looks like this one lost the filter that I had added to it. But okay. Okay. And let's go to this one and see Great time. Okay. **[00:24:00]** Okay. Now, good. Now, we're going to go back. I think this is pretty useful. So, another use case for this is like when you go and get a haircut, right? So, maybe you have different styles that you like to use. So, you can then just show them like, oh, this was my last haircut or the previous one or this one they did it right. This one they didn't. So here looks good but the second image has the guest my Carlo tag though it does not show it in this current view. I don't know what else what else should I just saw a tweet here that was funny. To replace programmers with AI, clients will need to accurately describe what they want. We're safe. That's that's sort of true. So, it's not and that was what we were talking about at the beginning, right? So, it's not it's not so much that these AIs will replace everyone because you still need to know what you want. And so a lot of times a software engineer, not just any developer, but as an engineer will understand and translate the customer's requirements or what they think they want into what they actually want and then propose a solution for it. And so if you can do that, then yeah, you can use these tools to just generate it. But but you still have to do that step. And if nobody does that step, whether it's the client or the engineer or the consultant or whoever, then you're not going to get anything useful, right? you still you still need to know for fortunately as you can see here these tools are getting quite good at at sort of inferring what you want uh which is very interesting but **[00:26:00]** uh what should we add I don't know what else I think I mean for now it's it's literally doing what what I need what I needed to do we want I I don't know should we should we uh yeah this looks good now let's Let's am running most of the time. I want to be able to from the main screen just ask it in natural language, what did I wear last time to the podcast or Uh, on the last podcast with Mike, prompt section for this, please. I don't know. This is taking a little far. I don't know if it's going to do it. I because it would it would need to yes of course the UI it could but but then it would need to embed some sort of um LLM feature into it or connect it to something. So I don't know if if it if it kind of got far but what did I wear to the podcast last time? So I clicked enter again. This is taking Oh, Okay. Um, would this connected to an LLM or how would it understand? And this is an interesting point like when you're doing something like this, obviously this is just a playground. We're not really doing it with it. But here it's it looks like it's just simulating it, right? It's saying the user is asking about a feature. Okay. Yeah, this is a good question about implementation. the context of I'm simulating the behavior but in a real application yes it would be connected to an LLM to understand natural language queries like LLM cloud GPT bars etc etc I'm demonstrating the UX how it work but in reality it would need an AI back end right so as you can see **[00:28:00]** since it is an LLM it can actually do the simulation of what that would be and this brings up another interesting point which is need something that's sort of disposable, these sort of interfaces that are just there while you use it and then can be thrown away would be very interesting because you don't need in this type of scenario consider this like a little artifact that it can make. It it has its own LM built in. So it could just do the processing directly on it and you don't need to build an API connection or an MCP connection to Chad GPT or Gemini or cloud. It could just do it within the thing. So if these tools actually come up with something like this that you can just use and keep there and use again and things like that or even host it for other people to use then it it becomes really powerful. So this this is I think I am satisfied with how it's how it's working. What do you think? So if you have any questions again throw them in the chat. Sometimes the chat here will will not let us know when it's like for example here Anton said the video feed is frozen. That's not cool. So let's see what that looks like on YouTube. So sometimes the the the chat messages don't come in through Riverside which is uncool. So I have to go into each of the platforms and check. So let's see if Where is it live? Is it frozen? No. Here I am. But thank you, Anton. Uh because maybe maybe it froze a little while ago. Yeah. Okay. Well, YouTube had such a bad delay. Anyway, that is what I wanted to show you today. **[00:30:00]** So, again, if you're a business owner and you're trying to see how all this works out, you can reach out. We'll happy to help. But it's there's so much you can do. There's so many things that are possible for a business to do either using the coding tools or using no code tools and just using these LMS to augment what you need to do. If you need to things like workflows for things that are just repeated tasks that you're doing day in day out. You're just constantly going into a spreadsheet, moving the data to the other spreadsheet, moving to another spreadsheet, putting it into QuickBooks, sending it to the accountant. All those things or most of those things can be automated and you can gain back a lot of your time. And this also applies for speed to lead. There's a lot of people in the services industry, home services for example, where you do get a lot of leads, but you can't get to them fast enough. So you can create little systems that automate this process. And we're doing this for a few clients where they're getting all their leads from like Angie and Thumbtac and things like that. Then we can process them, receive them in one central place and then start the outreach to each of these leads because remember they are leads. They're people that are they're asking for things, but maybe the practitioner doesn't have the time to get to them. So what we can build is a little interface that that brings them all in, gets the important information, then reaches out via email, text, and a a voice agent call that can then qualify the person, ask them, give them additional context, and even book the visit **[00:32:00]** directly on the phone without the the practitioner, the the professional doing that part of the work. Because at the end of the day, the person just wants to get their questions answers whether whether that's the that's what they need or what does the visit look like, how much it cost, etc. And so if the voice agent can do it 24/7 is perfect because it could just book the visits and then whoever is doing the work can just go and attend those visits and then now in real life talk with the customers, see what they need and help them. So really powerful stuff. We're in a new age really. There's there's so much these tools can do. It's more of more than just a little chat thing that can give you a summary or the weather. Really, there's there's a lot, especially in business. There's so much it can help you with. So, I hope you enjoyed this live stream and we'll see