Why & When You Should be Using Claude over ChatGPT

So these days there's really a lot of good and very useful AIs, aka LLMs out there. But across our team, the AI Advantage community, my professional network and the wider Internet, the consensus seems to be clear. There's really two of them that stand out. OpenAI's ChatGPT and Anthropics Claude. Right now, Sonnet 3.5 is the best model and both of these offer advanced features in forms of GPTs and projects. These are more specific ways of working with the models rather than just talking to it as you would to an assistant or a human. But this seemed to have caused a lot of confusion, particularly the Claude Claude projects feature, because it is very different from the GPTs feature inside of ChatGPT. These days I get many mails, comments and questions in the community around when to use what, when do I want to use ChatGPT, when do I want to use Claude, when do I use GPT, when do I use projects? And there's actually quite specific answers to all of these questions and in this video I'll present all of them to you. One more honorable mention should actually go to Gemini 1.5 Pro from Google. Their model has the largest context window and it has the added capability of you being able to upload videos, but that's a different topic. So without further ado, when do you use GPTs? When do you use projects? Let's start by looking at GPTs and their advantages and disadvantages as of today. Some of the advantages here are very clear. The tooling is just unmatched. You have the Data analysis tool, formerly known as the code Interpreter, which helps you analyze data by running Python code in a sandbox. This opens up a lot of opportunities with either data visualization or cleaning and all sorts of other little things that the packages available within Python in this environment can do for you. Then you obviously have the Image Generation tool with Dall E. This is a unique capability within ChatGPT that you do not have with Anthropic. As of now, you also have a browsing feature. It can go out into the Internet, retrieve up to five links, and use the content of those pages within its answer. It's not incredible, but it's good enough. A lot of times it works. Although it is a little hesitant to use this, you often have to specifically tell it, hey, go out there and browse the Internet, go to this URL and check it out. Nevertheless, it's not perfect, but it's there and you can use it. Here's another advantage that in my opinion is Maybe the biggest argument here, if you build GPTs versus projects, you can actually share them. And I don't mean just sharing the conversation, you can actually share the GPT and let somebody else use it. Now, with GPT4.0 being free to everybody, only a few messages, but still free, you could send the GPT to any one of your co workers or family members and they can simply check it out without having a paid plan. This can be very powerful if you solve a specific problem of the GPT that the other person might not know how to approach with AI. Another unique advantage of GPTs is that you have these conversational starters, aka prompt presets. Again, this is a feature that you want. If you set it up for somebody else, you don't want them to guess what kind of prompt they should run. You just want to set it up for them, share it with them, and they get the results. And then obviously you have actions which allows you to implement external API calls into your GPT. So you can do things like save the results of a conversation to a database or trigger some other application as soon as the GPT is done doing its thing. And then, last but not least, when it comes to advantages of GPTs over projects is you have the mobile app and it's really easy to switch context from your laptop to your mobile phone. It's actually really convenient. And if you have use cases that you might want to use on the go, it's great to have the mobility between the GPTs. Now, what are the disadvantages of GPTs? Well, they're kind of a reflection of the advantages of the projects, but let me briefly go through them and as mentioned previously, after going through these advantages and disadvantages, look at what concrete examples you might want to be using GPTs over Claude for or the other way around. So in the end we get more specific rather than talk about the features. But the features are important because to achieve a lot of results you do need to have knowledge of the features. So, disadvantages of GPTs? Well, first of all, a lot of people state, and I would agree, that generally anthropics models have a preferable tone of voice and style to their writing. It's just more human, more fluid. It doesn't have this chatgpt stink that so many people point out. And if we're honest, most people are kind of allergic to it at this point, at least within the AI bubble that we're in. It might be a subjective thing, but most people do agree on this. Another disadvantage of GPTs is that they have A limited context window in the Knowledge Base. But a few months ago they shipped a massive upgrade to the Knowledge Base feature. Now look, I don't dare to publicly state an exact number of how much context you can actually put in there, because it just varies in practice. While teaching GPT building in the AI Advantage community, I came up with this rule of up to 15 pages always works. So whether that's one PDF with 15 pages or it's 15 one page PDFs up to that amount, you're always safe and it includes everything. If you venture beyond that here and there, it starts losing pieces of the context and I wouldn't fully trust it. Now, can you upload 100 pages and is it going to work? Yes. Is it going to miss out on information here and there? Also, yes, this seems to be less of an issue with Claude, but again, this is a very subjective and intuitive thing that I'm communicating here. They have no official data on exactly how many tokens are considered within these documents, but let me just tell you from all the collective experience that I've soaked in so far, plot projects can take in way more context and it works more reliably. Both of them work, but if you have up to 15 pages, both of them work flawlessly. And then the Last disadvantage of GPTs is that you can always only have one chat. As you can see in a second here, projects are set up in a way where you can have multiple chats as you're working on the project. This will make a lot of sense when we talk about use cases, but before we do that, let's talk about one more thing, which is the advantages and disadvantages of cloud projects. Now these are kind of the opposites of the pluses and minuses we just talked about here. So let me go through these briefly. Cloud projects allows you to have multiple chats in one interface. As you're working on the projects, you can start new and new chats and you can keep everything in one place. It's sort of a folder for chats, something that I've been asking for and ChatGPT ever since it came out. It's one of the obvious features that OpenAI has refused to release so far, probably because they're more oriented towards this agentic feature of letting the LLM figure it out and building assistants that work for you, rather than a LLM that works with you. Next up, Claude is just way better at cool code, especially the Sonnet 3.5 model. This consensus just has been established at this point across the Internet. How did they do it? There's many theories out there. They might have trained the model on synthetic code, or there's some other black magic going on in the background. But in practice, if you're doing coding projects, Sonnet 3.5 is usually better, with the exception of longer outputs of GPT4O. So if you need to put out a script that is a little longer, GPT4O within ChatGPT has more tokens that it can output in one shot. That is the one advantage when it comes to coding. Okay, next up. Cloud projects also have better tone and style, as we pointed out. Again, might be subjective, but most people do agree it sounds more human. And then we have the longer context within the knowledge base. This, you work on your own project and that's it. There's no prompt presets with conversation starters. You cannot access external APIs and do things across the Internet or your own apps. With something like actions and GPTs, there's no image generation, there's no browsing, there's no data analysis tool. All right, so those are the features. A few more important notes for you once you're using these tools. So one really weird one, and I wish they changed this is if you use the search function with an anthropics claude, it doesn't look inside of the project, it only looks at the name and the description. So if you have a conversation and something comes up in there, you cannot search for that. As you accumulate more and more projects, more and more chats, it's really nice to be able to look into the chat. That's kind of the whole point of the search function. Just be aware with projects, it only looks at the names and descriptions. As of now, please change that anthropic. Maybe we don't need the knowledge base. I understand how that can be a lot of tokens, but the chats should be accessible. Secondly, in both features, you can kind of pin the most important ones. In GPT, it's called that. It's called pinning. And inside of Claude you can start either project or chats. Really convenient. I use it all the time. You should also be aware of the fact that you have custom instructions for both of these, which are incredibly useful. I've been saying it in every prompting related video on this channel, how great custom instructions are and how they allow you to personalize your experience for yourself. I mean, if you just look at all the Apple announcements, their whole thing is, hey, we might not have the best model, but we have your personal context and it's present within Every conversation. And that's why we can unlock all these incredible intuitive and easy to use features for not millions, but billions of consumers. And the whole thing is based upon your personal context. That's why I've been preaching setting up custom instructions with your personal context since around March 2023. They're super important, and if you give it your personal context, you can actually turn a normal chat into a personalized AI assistant. And with GPTs you can share it with other people. So you can empower your whole company, department, whatever it might be. Super powerful stuff. Use custom instructions. Okay. And last, before we get to use cases, that Claude actually doesn't offer a store, but it doesn't need it. It's projects. And usually projects are going to be a one off thing. Once you're done completing the project, you kind of move on. GPTs are more like an employee that you want to hire on a recurring basis. Now you might not need the employee for 8 hours a day, but maybe you need it for 30 minutes every single week. This is where GPTs shine. Whereas with projects, once you're done, you're going to archive them and probably not return because the project is done. So now let's get to the part of when you should be using projects and GPTs in concrete scenarios. Not just the features, not just things to be aware of. In what scenarios do I use them and what scenario do team members, community members use? Claude Projects? Well, as the name suggests, everything that is project based. Usually you're better served with CLAUDE projects, as long as you don't need any of the tooling inside of GPTs. And even more concretely, if you're writing something, Cloth Projects is excellent because you can give it examples of your writing just like this. You put in a few examples of previous emails that you wrote, you save it as a PDF, upload it to the CLAUDE Projects knowledge base, and then once you prompt, you can always reference that right in the style of name of the document PDF. And it just works super well. With the improved tone, the extensive knowledge base where you can upload a bunch of your writing to, you can just start writing all types of things that you need to get knocked out. Do you need new email drafts or some presentation, whatever it might be, as long as it's loosely related and you want to keep the style consistent. Projects are a fantastic way to do this because with Claude's models like Sonnet 3.5 or Opus, and between those two, you'll just have to test yourself, they're very similar for coding Sonnet is better, but for writing either Sonnet or Opus are fantastic. I think you're just going to be very happy with the results and you can keep the entire project within one chat. Very useful. But the next example is even better, and that is if you're creating a coding related project, if you're creating a game or a website and you have multiple code snippets that you don't all want to be housed in one page, well, just upload the various Python files to projects and it will be aware of your entire application. And now if you're prompting something new, you're doing it within the context of your entire application and not just the one code snippet or just the conversation you had. So far it's aware of the entire project. Huge. Now, could you do this with GPTs too? Yes, but again, coding abilities within PROPICS model are just better. So using coding projects for this is a no brainer. Now, yes, as I mentioned before, maybe you need the longer output of GPT4O and you can generate some of the code in there if you need a longer script, but then you can import it to CLAUDE projects, just add the Python file to the knowledge base and you take advantage of the long context awareness of CLAUDE to complete your projects. Although some of the code might have come from ChatGPT. So as you can see, to get the most out of these models today, you need to be aware of the multiple advantages and disadvantages. And ideally you mix and match them. And then also cloud projects are just generally speaking, better for anything where you're going to have five or more chats, three or more chats, and you're not exactly sure what they're going to look like. With GPTs, it's more like a laid out track that you're going to be following. That's what really makes sense within that interface. But with projects it's more exploratory, it's more like, okay, I'm starting this project, here's a bunch of context, here's some of my writing, here's some Python files that I might have, and now let me talk to you and let me figure out what else we need to add here. Whereas if you build a GPT, you already know what the use case is, you already know what you're going to be doing. So that's cloud projects. And when do you want to use GPTs? Well, a very clear one is if you want to build a solution for somebody else, if you want to share a AI application or a prompt with a colleague there's no better ways to do it today than with GPTs. You can customize it to their context with the customer instructions. You can add useful prompts to them in the conversation studies. You can even add an action, and this is the most common one that I personally use that saves it to a database. So you bring it outside of ChatGPT without having to do anything. Just simply add a keyboard shortcut that engages the action and you can save it to a notion database. And your counterpart doesn't need to know anything about prompt engineering, API calls, or even how to use it because you set it all up. The instructions are there, the prompt presets are there, the action is there. They just go in and once they're done, they press one key and everything that you need gets saved to a database. Now look, this is quite complex and if you're interested in learning this, yes, I do teach you. This is one of the learning paths within our AI Advantage community. But again, we give the base presets out for free. So if you have enough knowledge and if you want to just customize these and figure all of this out by yourself, you can absolutely do that. This channel is here to support you on that journey too. Another use case for GPTs is everything that has to do with assistants, tutors or coaches. It might be a base model, it might be the system prompt, I'm not exactly sure, but when it comes to tutoring and actually assisting you with your work, GPTs are just so much better. The entire ethos behind GPT4.0 is the model trying to get your work done first for you or with you. It's very conversational, it's very practical. And when it comes to something like an assistant or a tutor, GPTs are just the perfect tooling for that. Actually, with my free GPT Builder GPT, one of the best use cases is building little custom tutors for whatever you might be learning. I have a separate video that you can check out. The free GPT that builds GPTs for you in the description below. And another reason to use GPTs is obviously the tooling. Data analysis, image generation and web browsing is just something you don't get with Anthropics cloth today. I'm sure it'll come over time, but as of today, if you want these, well, got to use GPT4All. And then last but not least, I would just generalize and say GPTs are just better if you have a repetitive task that is super specific. And this is the reason why the store didn't work so well because the store tries to give you generalized solutions and the best GPTs are the ones that are custom to your very specific context. It's the reason why I think that the Apple integrations will actually work really well for all the users there because they have your personal context and the GPTs from the store don't. They're generic by definition, but Everybody who integrates AI into their workflows and actually builds custom GPTs finds that yeah, of course it is useful to have a brainstorming buddy for every single YouTube video that is aware of the last 10 uploads, all the titles and thumbnails I've ever used, and has a keyboard shortcut menu of various brainstorming techniques that I apply every single time when I create new video ideas. Of course that's useful, but that's probably not useful to you. That's a GPT that I built for myself and that I use all the time, but that's because it has all my personal data in context and that's why it works so well. This is where GPT shines versus projects are exactly that. They're a one off project where you might not be sure where the journey leads yet and once you're done with it, going to archive it and that's the end of it. All right, I hope this video was helpful to you. All the resources are in the description below. Subscribe to the newsletter for the free GPT assistants that I mentioned. And if you want to learn even more about CLAUDE and see some of the examples that you could build with it, like turning PDFs into websites, or turning screens into animations, or just creating games from scratch, check out this video where I pointed out 21 use cases for Claude 3.5 sonnets. I'll see you there.