🤖 Why you need to check out Claude's Projects

I have been using Anthropic’s Claude AI for a few months now and I find myself using it more than any other product out there. I have found Claude 3.5 to be measurably better for the kind of tasks that I work on. I will dive into a lot of those use cases in future issues but today I want to focus on one feature that I find really useful.

The Problem

You heard about the amazing problems these Large Language Models are trying solve and you want to see if it can help you with your next project. Let’s say for example you were looking for assistance with writing your next blog post. You head over to Chat GPT or Claude AI and enter the following prompt:

If you have done something like this before you know the results can be less than desirable. This is where the art of Prompt Engineering comes into play. This is the practice of designing and refining input instructions for AI language models to optimize their outputs for specific tasks or desired outcomes. In simpler terms, learning how to effectively communicate with the LLM.

This is where you begin to iterate until you get an output that you’re happy with. The first thing you can do is to give it more information about what the blog post should contain and the length of the post.

This is starting to produce a better result but it still isn’t exactly what you are looking for. Next you decide to give it more direction on the output like style, tone or format.

This would now give us a much better result than our first attempt. This is just the beginning of prompt engineering. There are guidelines, tips and tricks that we could continue to iterate on to get better results. We don’t have time in this ByteSized issue but if that is something you’re interested in learning more about let me know.

Introducing Claude Projects

This was a single blog post. Can you imagine having to do this for every single blog post you wrote? This is where projects allow you to take anything you will repeat in the future and templatize it. This lets you build your own customized chatbots.

Before we get into to how to use I thought I would take a moment to talk about the different categorization of roles in an LLM. A message is structured and interpreted in many LLMs by 4 primary roles:

• User role: Input from the human user or client application.

• System role: Sets overall context and instructions for the AI's behavior.

• Assistant role: Represents the AI model's responses.

• Function role: Output from external functions or tools the AI can use.

For the purpose of this conversation we will focus on the first 2. The user role is the prompts that we were iterating on above. The system role will be responsible for setting the overall context. In the case of our AI blog post writer this could be purpose, style, tone and format.

Getting Started with Claude Projects

To get started with Claude Projects head over to https://claude.ai. You will need to sign up for a Professional Plan to take advantage of this feature but in my opinion it is well worth it. In the left hand side bar you can click on projects and create new project.

From there you can give your project a name and a description. This is just for you this is not any information that will be used in your user or system messages.

This will be your home for every blog post you write for this project. If you look on the right side under project knowledge this is where you can really fine tune the project.

If you click “Set custom instructions” button this is where you can set the System message for this project. What I love about this is that the custom instructions are per project so we can really refine the overall context on a per project basis.

You can also use variables within the custom instructions and then set that variable in each individual chat. Instead of saying write a blog post on starting a nonprofit we can say “Write a blog post on ” and then replace it with our current topic in each chat.

Under project knowledge you can also upload multiple documents as its knowledge base and add your own specific instructions. This could be examples of your writing to adhere to, formatting options and more. This will be knowledge for every chat you initiate from the project. Each project includes a 200k context window and you can include files, documents and code.

Finally, when it comes time to write a new blog post you can also add documents specific to this chat. Let’s say I was writing a new blog post on Prompt Engineering Best Practices. For this one I will include an updated version of Open AI’s Prompt Engineering guide.

If you want to learn about some prompt engineering best practices for Claude you can check out their guide here.

Getting Creative with Projects

What we just saw was how to setup a project for a very specific use case. There are times where you want your prompts to use some custom instructions but they aren’t as specific as the project we created above. Alex Albert has a great Twitter thread where he goes through some creative ways he is using projects in his day to day workflow.

I have been using this approach to setup general projects for health, fitness, writing, coding and more.

Conclusion

Think of Claude Projects as your own personal team of AI helpers. You can set them up to tackle all sorts of tasks, from cranking out blog posts to diving into tricky topics. It's like having a bunch of smart friends who know exactly how you like things done – pretty cool, right?