New ChatGPT vs. Claude Projects (Favorite Feature)

All right, in today's video we'll be comparing two of the biggest players in the AI platform game ChatGPT and Anthropics. Claude. ChatGPT just recently included a feature that has been requested forever Projects. This allows you to organize multiple chats by purpose. Claude Synthropic had this already, but there are several differences between Claude's project and the ChatGPT project, and that's really going to be the focus of this video. Plus, I'll show you a few ways I and the team use projects in our very own workflows and we'll have a quick chat about which one of these models you would prefer in what scenario. So think of this as an up to date Claude vs ChatGPT comparison with a focus of one of my favorite features projects. So I think this might be quite intuitive to most people, but I want to start out by creating a project inside of both Claude and ChatGPT to show you the difference. Then we'll proceed to talk about use cases of this and we'll round this out with a little comparison battle on seven different categories that I established here. And in the end, hopefully you'll have all the information you need to make an informed decision on which one to use or which one to prefer. Prefer in what situation. So we'll be using a paid plan on both my OpenAI and Anthropic accounts here, the $20 tier that is. And unfortunately, projects are something that are exclusive to the paid plans. Now let's get into it. All right, let's start out in Claude as they were the first one to release this. So I can simply access this inside of my account when I click on projects. And as you can see, I have a few of them active here and actually actively use half of these regularly, but didn't create any new ones recently. If I wanted to do that, I could create Project Project like so give it a name. In this case I'm going to call this email shortener and once I add a name and a little description, I'm ready to go. So this is how you set up the project and we'll talk about the different features and how they compare here in a second. In ChatGPT, it's very similar. It's kind of the same button here in the sidebar you just say new project. I'll also call this email shortener. No real description needed. And any second it should have my project right here. All right, so as you can see, this is very, very similar. The basic functionalities, you can add files that will be stored in so called embeddings, effectively extending the context window of your your chats. This way you can add way more context than you could do just for chat. And then you have custom instructions that are specific to this project. These supersede any other custom instructions you might have. And we'll talk about that soon. And then you can see all the different chats that you will engage with in here. And also you get the model Choice here with ChatGPT. Same thing here in Claude. Between the anthropic models, custom instructions would be over here and the project knowledge can be added up here. So now that we talked about setup, let's talk about use cases and then a more detailed comparison on what the these two feature. On the surface they might look the same, but there are some nuances that I think you should be aware of. Okay, so what can you use this for? Well, first of all, this is a really good interface to transfer your own style onto the writing. We talked about this before. You can specify different aspects of your style like the tone, voice, or even specific words or phrases that you would want to include or avoid. And what works best is including multiple examples of your writing. You would put this into the custom instruction. I can show you one of my projects here where I did exactly that. Igor's email writer. If I open this up, I just include a simple prompt like this and then included free emails in this exact format, format email, one column, then include the email. Oh, it's actually five emails. You can see this is email number five. And then it will write in this exact style every single time. Now in the news you can use episodes. I showed you the style function which I really like. I actually keep returning to this here. It actually creates your style for you. Maybe an intermediate tip that I already shared is that if you create your styles in here and you provide it with writing examples of multiple emails that you wrote yourself emails of the same type, this thing will go ahead and do the same thing that it did here with with my examples. And if you go into here and say edit style manually, you will see a very detailed description of what your style is like. So if you don't want to formulate it yourself, just upload multiple examples to cloth styles and then you can use all of this. Ideally providing it with five examples, not just one. Now if you include this inside of your instructions, you will get the same effect as if you include it here in the styles and then everything in the project will reflect that one style that you now described and provided multiple examples of. In ChatGPT, it's the same thing. You would just put it here inside of the instructions, and now you have a custom writer. This is one of the best uses of styles. Nevertheless, it's not perfect. As I also mentioned, the best way to infuse your style into a large language model is fine tuning. But that's another topic for another video. We're actually building a little course around that in the community coming in the first quarter of 2025. So that right there is styles, and that's really good. But another thing that this can do is maintain specific structures of something that you need to write regularly. And I know that my examples here are a bit more writing focused, but that's what I really like to use these projects for. I'll point out some more examples that I've seen that I don't personally use regularly in a second here. But what I mean by this structured writing is the same thing. I provided a little bit of context and a task, and then I give it free example descriptions and for example, event descriptions in a community, they always follow a specific structure. And I just gave it free handwritten descriptions here and then it can replicate those. And all I need to do is say something like, this is an event. Write me a description for an event titled LLM Foundations ChatGPT for Beginners, and it will flesh this out in exactly the format that I want to see if I want to do this for a new event. And I wanted to follow that exact template a few moments later. All right, after giving in a few more details, I can actually mix in my tech mentor style here and hit enter. And this will generate the event description in the perfect format in my voice. And there you go. That's a great starting point for event description. A good example of something that maybe doesn't need to be handwritten every single time. It just really needs to provide the info. Actually, to be fair, I'm not even sure I like the fact that I mixed in my style here. Let me just do this with the default one as provided in examples, and I have something that is usable. So those are just two writing examples. But another thing that this is regularly used for is learning about a topic. So some people like to use NotebookLM for this use case as it's really good at inserting sources and YouTube video links, and then it can generate these podcast style summaries, something you don't have in here. But if you're looking to use LLM to interact with files and maybe you're researching a certain topic or you're learning a new skill. Well, this can be a fantastic partner where you can upload the different documents, the different PDFs, the different screenshots of the document, tutorials or flow charts you have into a chat in your project and you can start multiple projects exploring multiple avenues. Learning coding is one of the best examples here. It's just super powerful as you can give it different pieces of code that you wrote and then you can always reference back, return to older chats and talk about them and refresh your mind on what you've done before. This is really good for learning any skill, but As I said, NotebookLM might be a good alternative here for coding concretely, I would recommend projects. That's kind of the last use case I want to provide you with here if you're trying to figure something out with code. Well, yes, you could set the custom instructions up in a way that provides context on where you're at currently and what your goal is with this project. And you could add the code that you have in forms of uploaded content. And every time you start a new chat you can have a new conversation around how to do things, how to approach them, what the next steps are. Now obviously this is also possible inside of something like a specific IDE that is AI powered, like Cursor or Windsurf, but this is definitely an alternative. Actually, one more use case that I recently have seen comes to mind that I want to add here, and that's basically business specific tasks. If there's maybe a manual or SOP to doing something, you upload that to the project knowledge here, set custom instructions on your goals and maybe some limitations that you have, and then you can use a specific project every time that topic comes up again or that process within a business comes up again. You don't need to reinfuse it with all the context every time. You just have it inside of the project and return to that when necessary. It's hard to believe that 2024 is already coming to an end. Man, this year has been an absolute whirlwind for me and I'm not big on New Year's resolutions, but I do like to reflect on the year that happens and think about what I would like to do more of in the next year. I suppose that's sort of a New Year's resolution, but whatever. The one thing that I consistently reflect on positively is building skills. It has been at the foundation of my life and also at the foundation of this channel. And never in my life have I regretted investing the time in learning something new, I highly recommend you do the same. And one of the best ways to do that is with the sponsor of today's video, brilliant.org because brilliant is an online learning platform focused on interactive education in math, science, data analysis, programing, and so much more. If you've ever felt stuck just watching lectures and tutorials without anything truly clicking, Brilliant's hands on lessons are exactly what you need. They get you directly involved with the material, going way beyond just passively consuming. All of Brilliant's lessons and learning paths are designed to get you experimenting, making mistakes and having these, ah, I get it now type of moments. Just like I've been saying on the show all year, you'll get the best results by not just watching, but by interacting with some of these tools that we show off here. And that's exactly what Brilliant does best. So whether you want to level up in data analysis or enhance your knowledge of AI tools or Transformers, or you want to dive into a foundational subject like computer science before jumping into something like machine learning, Brilliant's got you covered. So head on over to the link in the description and start exploring Brilliant for free. And if you decide to keep going with it by using the link in the Description, you'll get 20% of your annual subscription. It's a great way to get yourself set up for a successful 2025. And once you build these skills that will last a lifetime, who knows where your journey will take you next? All right, now that we talk about different use cases, and by the way, if you have more, please leave a comment below. This is a fantastic way of sharing it between the community on this channel. But now let me take you for a little comparison of the different features and differences between Claude and ChatGPT projects, as that is most probably why you click this video. So here are the seven categories I'll be evaluating this on. Collaboration context, window model options, features and tools, custom instructions, platform support, and last but certainly not least, privacy. All right, let's just start from the top. Collaboration. So what collaboration options do we have within ChatGPT projects versus Claude projects? Well, there's a massive difference here because ChatGPT projects are built for the individual. As of late December 2024, ChatGPT projects do not allow you to share this within the team team, which I find a little curious as this would be the ultimate collaboration feature. I guess this decision has to do with the context. Len for I don't even know Claude if you're on a team Splan, on the other hand, Absolutely lets you collaborate, and that's one of the biggest upsides of CLAUDE projects. Now, it's important to note here that instead of Claude, you can only share snapshots. So it's not like you have an interactive chat where multiple people can contribute. This is what I would wish for, but none of these two actually offer that today. So inside of claude, multiple people can be inside of a project and share the context and the custom instructions, and then they can share snapshots of these chats, which are sort of like a screenshot of the chat at one point in time. ChatGPT projects are just for individuals. Therefore, in this category, I give the point to Claude. All right, next up, context. And there's a little bit of fog around this topic, but I'll do my very best to clarify what has been communicated here. CLAUDE has been quite clear. 200,000 tokens of context that spans across both the chat input, output, custom instructions, and the project and the files that are added. ChatGPT has not clearly specified how large the context window is, but just like ChatGPT4 and ChatGPT, I would assume, and also from the usage so far, it seems that there is 128,000 tokens of context, which honestly, for most use cases is totally fine. I think the most interesting part in the discussion around the context is how it processes the files, and in both cases it creates embeddings from the files that you add here to the file upload functions. In both of these project interfaces, what that means is that it doesn't store them verbatim, but it divides them into little chunks, and then every time you send a message, it looks across those chunks and then pulls out the ones that seem relevant to the conversation. This way it can store way more than any conversation could ever hold. So for anything that is super, super long context projects in both, these are the user friendliest way to do this. But nevertheless, CLAUDE has the bigger context window with 200,000 tokens. So Claude gives you more context. But as I said, if you use this upload feature, it will minimize them anyway and then the calculation happens in a different way than we're used to. Inside of ChatGPT and ChatGPT, you can just copy paste the entire conversation, throw it into a tokenizer, and you will know exactly how many tokens that is. With uploads is a bit more convoluted, as it depends on the chunk size and things like that that it automatically determines in the background, we don't even get to see the technical aspect of it. So even though CLAUDE has technically the longer context window, I think the project knowledge feature kind of levels the playing field here as that is the main repository of context and both of these work. There's no specific research or info on how many tokens exactly you can put in there in this category. I would give both these a tie and an unreasonable test that we ran this through. We couldn't hit the context limit properly. If somebody knows even better than that, please leave a comment below. For me, this is a tie height category as both seem to work equally as well in the context department. All right, next up we have model options you can switch right here and this one is quite simple. You have GPT4O and 01 and then if you're using Claude you have 3.5 Sonnet and Opus, which has become quite niche. Some people like it for writing, but most people will just default to Opus. And this opens up the discussion of hey, which model is better for what? I think the consensus opinion is that hey, on code generation, Claude 3.5 sonnet is really excellent, but the new 01 actually beats it on at least output length and many people even claim that the code quality coming out of 01 and O1 Pro are even higher. Oh, by the way, that makes me want to check if I can use O1 Pro inside of projects if I'm on a Pro account. Okay, just quickly reloged and yeah, indeed I can use 01 Pro inside of this project. So I think with that, clearly this point has to go to ChatGPT as you have bigger model choice, you have the reasoning models between 01 and 01 Pro. You have GPT4 as the general purpose model. I think if it was just GPT4O versus Claude 3.5 sonnet, I'd clearly give this to 3.5 sonnet for the quality of the writing and the code generation. Those are the two big things that it really excels at. And also and those two, as I pointed out in the use cases, cover a lot of the things you will want to do with projects, although I'm sure you could get very creative here. But as this includes the reasoning models and also the O1 Pro mode for people on a Pro account in terms of model choice and capabilities. This model option category has to go to OpenAI. All right, next up, features and tools. This one is also quite simple as Claude is quite limited in terms of the tooling when compared to ChatGPT. So with GPT4O you can use various features like image generation with Dall E Something that Claude doesn't have, or the Canvas feature, something that Claude also doesn't have that allows you to interactively write and annotate pieces of text or code. I really like this feature, by the way. You can also upload files and images that's available in both of these. You can also go directly from Google Drive, which is available in both of these. You have a data analysis tool in both of them now, so that's kind of a tie. This one is hidden in ChatGPT. It's just under the hood. OpenAI does have the ability to search the web, not ChatGPT search, but the search functionality are slightly different. CLAUDE does not have this by default, although it could be included for Claude's MCP open source framework that allows you to include external tools. This is something a bit more advanced though. I just wanted to put pointed out. I ran a recent community event on installing this and there you can add things like web search or the ability for it to manipulate files on your desktop. The example that I was showing off, it can just reorganize your entire desktop or organize entire hard drives or folders for you. Only downside is every few minutes you need to tell it to continue. So you can add this tooling into Claude, allowing for more customizability. But this is more of an advanced technique. I think I have to judge based off what's possible in the default version. And in this case ChatGPT just wins out with all the extra tools as I just named them. So point to ChatGPT here. But I want to add the honorable mention of Claude artifacts that I still love. ChatGPT canvas cannot just write but also run Python code and then debug it. You can do the same in artifacts, but artifacts have the upside of the fact that it can generate HTML code and display it right away. So you can create front ends and websites really well. Something that you cannot do inside of ChatGPT. But I think all the other features still make it that ChatGPT wins this category. All right, next up we have custom instructions. One of my favorite ones, the ability to go one layer up and alter the behavior of the LLM in all interactions. So both ChatGPT and Claude include custom instructions in the settings of the profile in Claude, it's here in chatgpt, it's here under personalization. And here you could add various things. But it's important to note that if you have these set up for your account and then you're using projects, the project wants supersede what's inside of your profile custom instructions. So it's really the same functionality here. It just happens on a project level rather than an account wide level, making these two equal. I wanted to add one note and that is the fact that if you have any memories set up inside of your ChatGPT account, I know that this is a popular feature amongst many people. These memories will apply to your project chats, so keep that in mind every time. If you have this memory featured on and it captures various things about you. This was an example video I did a while back. I guess I could delete this memory. Gotta make sure that you manage this and that you're aware of what's in here because all of this information will be now infused into every new chat I open in my project, which in case of this one might be a good thing. In case of this one, maybe I don't want to work on something YouTube content creation related. Maybe this project is for something entirely different and then these memories would ruin it. So I personally keep my memories off most of the time and I vary between different custom instructions and in this case I will start the process of migrating a lot of my custom instructions and prompts to various projects in my main ChatGPT account as this will allow me to really easily context switch between different purposes. So if you follow this channel for a while and you have your own sets of custom instructions set up as I showed you before, you can create multiple projects with them here in your instructions and then you can really easily switch. Now this is the feature that we wanted since forever thanks OpenAI and also thanks Claude for having this even earlier than them equal functionality point to both of them. At this point I just want to take a second to give you some extra info or where to find some prompts or custom instructions that you could put in here. Now we talked about them, you might be interested in what you could even put in there. And we have this free resource that comes with signing up to our newsletter that we updated for 2025. We call it 2025 AI prompt library now. And essentially this includes some of the most used and most interesting prompts. And I would just like to quickly highlight two ways we have here at the AI Advantage where you could get these for free. One of them is something we call the Advantage Resource Compass. In under one minute you fill out four questions and you get a customized set of prompts for your interest, a resource and a course recommendation that could help you along your way. That's one way to get prompts and the second one is Signing up to our free newsletter. We just updated this for 2025. All the most useful prompts for the various categories of use cases we cover, like education, business, and many more. So for example, if you're in education, you want to see what prompts are really useful. Here's a compilation of all the different quiz results so you don't have to take this quiz a dozen times and get the different results. You can just look at this one template. And some of these prompts are rather advanced, but these are the ones that we found that work and stood the test of time. And you could easily take some of these, customize them slightly, and build your own projects around them. Links to everything I talk about in this video can be found in the video description below, but that'll be my recommendation. If you want to come up with new projects, just get this free file, look at some of these categories that might fit your interest, and pick some simple or more advanced prompts that suit your needs. All right, bringing me to the last two points. Platform support. This will be a very simple one. Both of these are accessible for the web interface, equal on that front, and Both Claude and ChatGPT have Windows and macOS apps. But the ability to create and edit projects is only accessible inside of ChatGPT. Web and windows for macOS like me, right here on your own mobile, you cannot create or edit projects in there, you can only use them. CLAUDE has the full support across all their apps, so it's a minor advantage, but this point has to go to cloud projects. Onto the final point, which is privacy. And look, this is a bit of a tricky one. Obviously if you're using an online model and you're sending off your data to some data center that is owned by one company, rented by another company like OpenAI, you're entering more of a risk than running a local model that doesn't even send things anywhere it doesn't need the Internet. Of course, given that the privacy policies do differ slightly, and I think CLAUDE has made it very clear that their main focus is privacy and safety. That is their biggest thing. I mean, it comes at the price of the models being a bit more lobotomized in certain aspects. But CLAUDE is kind of the leader in terms of, hey, your files are safe with us. That's their entire company ethos. And it also reflects in the privacy policies. For example, they do explicitly state that data and chats will not be used to train models without user consent. ChatGPT Projects did not mention this on the launch, but if you're subscribed to the team's plan all of the data will be excluded from the training and all the personal plans you have to manually opt out which you can do by going to settings data controls and then here improve the model for everyone you need to turn off otherwise they can use the data to improve the models or train other models on so as Anthropic's main focus is privacy and security and it comes with this model training feature turned off by default I got to get this point to Claude which means that Claude is actually the winner here with five points versus four. Now obviously this is super close and if you already have a subscription of a preference in terms of platform this means that you you should probably just stick to what you have. Nevertheless I hope this video was helpful for you personally or in a business setting. If you enjoyed this don't forget to leave a like and I will see you very soon. And with all that being said I hope you have a wonderful day.